kerull leu Uy
Contents
1. More recently there have been animation systems such as that of Catmull 1 which will generate halftone cartoons of arbitrary complexity from an easily understood set of commands Designed for cartoonists with little or no computer experience it only manipulates picture features picture processing is not implemented There are also a wide range of graphics packages of which the most popular is probably GRAFPAC These provide a library of subroutines which draw points lines circles etc on graphics displays Programs using such packages by their very nature are generally not interactive at least regarding the specification of their behavior Changing their operation requires programming the appropriate subroutine calls recompiling and reloading Furthermore the host language usually FORTRAN is generally not very well suited for picture manipulation There are also interactive graphics systems designed for special purposes such as curve fitting While these perform 1 Catmull Edwin A System for Computer Generated Movies Proceedings of the Association for Computing Machinery August 1972 Annual Conference p 422 11 well in their area of intended use they are not suited for development work One example of a picture processing system is BUGSYS 1 which permits fairly extensive examination and change of previously scanned pictures on a pixel by pixel basis It is however difficult to make significant modificat2. a long time Hence this requirement had a strong influence on the language A PIPO command is a specification of a series of operations to be performed in order to produce an output pixel The command is iterated over a specified portion of the picture producing one pixel of the new picture with each iteration For operands a command may use local pixels constants or values in a separate array Use of local pixels is done through reference to a matrix whose elements are numbered as follows This matrix is overlaid on the upper left hand corner of the picture and shifted over between each iteration until it 22 has covered the entire picture The pixel at position 0 0 is the current pixel Its absolute coordinates in the input picture are equal to the coordinates of the output pixel that will result from the iteration In other words if the operation copies pixel 0 0 to the output on each iteration the new picture will be identical to the old The IPMS command to specify this operation is just DO 0 0 The DO command translates its arguments into a PIPO program and executes it over the picture The pictures to be used for input and output and the portion of each picture the subpicture actually used are specified separately If 1 1 had been specified instead the output picture would have been shifted diagonally upward and to the left Note that this notation does not permit specification of translation down or
3. 2 9 v9 s onuboe s wee e a e sq eee 0 0 570 0 A I trod ction s o o4 os vo a oe Pei i eee ea B Overview o9 9 sii slavere re o ww eke wie le clea See Gx The DO and FILL Commands oie ee ace ee ieee ROS eee ee The DO Command ccccccccrccccevccrvseressseeseseseecs A IntroductioOn sssessseeseseosessossenssnnnnsaasnaebnn ana B Implementation of the DO Command ssssecesssesossesesesos C Reducing I O Overhead sscsssessssssnsonnns onn D Extensions to the Languag sssssessessesensnosnonna ore Screening and the FILL Command cccescccccceveses A Introdu uction is d as RP ev eo Ra IH 7o bare y eda B Producing Sereened Pictures on the Phototypesetter CONC LUST OWS acd ok x o9 Sew o a oe wu ar ecu m ble REF UR Re Page 5 5 10 14 14 Chapter I Introduction Computers have been used to generate and display pictures ever since the first line printer was used to make pinups It is only within the past few years however that they have begun to be employed widely in commercial picture production Similarly typesetters have traditionally been used only to produce text pictures were handled by a combination of photography and pasteups and only recently has there been a move towards processing pictures by computer control of the typesetters themselves An example of this changeover is Yellow Pages directory production Its long term goal is a system in which the entire directory pictures as well as text is
4. a coordinate specification of the form m or m n which it would translate into an appropriate offset indexed by a register which will later hold the address of the current pixel If a constant were specified as the operand to the operator the table matcher would find a match for which would have the instruction add immediate and type immediate The translator would expect a number which it would insert into the instruction Note that the is effectively part of the opcode rather than the operand a difference which is invisible to the user One operator which requires special assistance is the operator which divides the current value by its operand The SEL requires that the dividend be 6 bits long spread over two registers Hence this operator is flagged so that an extend sign instruction is put out first extending the current value into the adjacent register Another group of operators requiring special handling are the ones that implement conditionals IF ELSE and FI The IF operator actually consists of the IF and its one or two character condition name The operator is translated into the appropriate conditional branch However since this is a branch forward the translator does not yet know where it is branching to So it stacks the location of the IF When it finds an ELSE or FI it unstacks the location and fills it in 34 with the address of the next instruction ELSE is of course an unco
5. be screened and where it did not the artist could draw curves to indicate the areas explicitly At the console specifying any point within the curve would indicate that the entire area should be filled To specify the point it was decided that the TV and graphic tablet would be the most convenient input device Hence a command was written to display a cursor on the TV and read from 29 the tablet The coordinates obtained when the user depressed the stylus were stored as the values of abbreviations where any other command could reference them A second command was written to fill an arbitrary closed curve with points of a given color Given the user s specification of any point within the area to be filled the color of the curve to be followed and the color to be filled in the command generated an output file containing the filler There were several requirements for the fill routine It had to be reasonably fast despite the fact that individual scan lines of the picture containing the curve would be on a slow random access device The routine also had to avoid requiring main memory proportional to the area to be filled as it might be quite large and the routine would eventually be implemented on a minicomputer Achieving these requirements took some work The command had to keep track of the portion of the area already filled in The simplest way to keep this information was to use the filler picture being built up Refinements w
6. has been obtained however it is retained until the end of the session or longer if the user saves it and restores it later One other important job of the picture I O routines is handling the TV Adjacent scan lines in the TV are located half a frame apart The amount of TV buffer memory needed to hold one frame varies depending on the size of a TV pixel which can range from 1 to 4 bits Smaller pixel sizes permit more than one frame of data to be held in the TV separate commands select the frame displayed Furthermore if the TV is set to a pixel size of 3 bits there aren t even an integral number of words per scan line All these complications are handled by the picture I O routines a simple flag indicates whether the routines should address the raw file like a TV Since the picture I O routines do not actually know whether they are driving the TV or not the user can prepare TV format files ahead of time and display them quickly by doing a direct copy The DO and FILL Commands IPMS presently implements 12 commands directly that is as FORTRAN subroutines Several more commands are actually abbreviations Two of these commands have already been discussed SET and SAVE Some other commands perform minor functions such as setting the pixel size or colors of the TV These are not important to understanding IPMS However 20 the DO and FILL commands are important enough to be described in some detail The first command
7. is the DO command It executes a specification of a local position invariant picture operation over a specified portion of a picture The result of the command is a new picture the input picture is unmodified under normal circumstances The specification is in a language called PIPO for Position Invariant Picture Operations PIPO is described in detail in the next chapter the following description is meant only as an overview PIPO provides a convenient interactive way of specifying a large class of local position invariant operations on a picture including arithmetic and logical operations among pixels conditional operations and table lookup The PIPO language was designed to meet the following objectives 1 It should permit easy natural expression of picture operations 2 It should be compact so that it will be easy for users to specify an operation at the keyboard 3 It should be simple to understand 4 It should be straightforward to translate a PIPO command into very efficient code 21 The last objective was particularly important in view of the tight timing requirements imposed by an interactive system Although IPMS does not claim to be a real time picture processing system typical PIPO commands should not take more than about 15 seconds to execute or the system is no longer truly interactive But an operation on all the elements of a 512x508 TV picture must be very fast in order not to take
8. its arguments into SEL machine code and executes the result on the picture Both the translation and the resulting code are quite fast indeed the code produced is nearly as efficient as hand coded assembler would be The translator achieves this efficiency through a combination of careful language design and clever implementation The PIPO language is actually close to the machine language of the SEL in many respects in fact there is very nearly a one to one correspondence between PIPO operators and the resulting machine instructions The translation process consists of matching the first four characters of the string from which all blanks have been removed against the list of operators Some operators are shorter than others in which case less than four characters of the input string are actually compared against In any event further action is dependent on the table entries associated with that operator These entries are 1 the machine instruction to be used 2 a value indicating the type of operand and 3 an optional set of flags for the operator For example for the operator the instruction obtained would be add byte 1 and its type would be memref 1 This is not the actual name of the instruction in SEL assembler throughout this discussion long mnemonic names will be used rather than the SEL names 33 meaning that a matrix reference was expected to follow in the input string The translator would expect
9. s equipment 13 Chapter II IPMS Structure Introduction As discussed in the previous chapter the picture processing routines of main interest at the time the system was begun included the following 1 2 3 Screening specified areas of a picture This is actually a two step operation specifying the area to be screened and converting the specification into an appropriate file of phototypesetter commands Drop Shadowing adding simulated shadows to objects In particular it was desired to generate drop shadowed characters Outlining essentially detecting the edges in two level black and white pictures This included both generating tracings around the edges of the objects and hollowing out the insides In order to develop the routines required a framework was designed in which to carry on the necessary development quickly and easily The framework had to provide the ability to run individual routines whose precise nature could not be foreseen when the framework was designed routines whose arguments might differ considerably both from routine to routine and from call to call of the same routine A command processor 14 was called for that is a routine whose sole job was to accept user instructions describing routines to be executed and execute them in the appropriate environment Another requirement of the system was that it facilitate the orthogonal command set phi
10. the leftmost white point on the line above push its coordinates find the leftmost white point on the line below push its coordinates fill in the current line pop the last point pushed assign to xl yl until there are no more points to pop 44 Here is an example of the operation of this algorithm Consider the following picture in which an at sign 8 is used to represent a black point and O P Q and S represent white points 00008880 o0eaGeaqoeo 80080080 eP000080 090000 e0 00000000 The algorithm is started at point S It scans to the left on the current line until it reaches the edge Then it scans right It immediately finds point P and pushes its coordinates Then it finds point Q at which time it pushes its coordinates Finally it reaches the end of the current line It fills it in The picture now looks like 00008880 08908Q080 20800008 e0 eP0000e0 00000 ee0 00000000 The coordinates of Q are popped off the stack and the process is repeated This time there are no points pushed as neither the line above nor the line below is white for any part of its length Finally the coordinates of P are popped off and its line is filled in This algorithm will work on any horizontally convex curve that is any curve satisfying the property that for 45 any two points on the same horizontal line a line drawn between them is contained entirely within the curve This can be seen from the fact that in order fo
11. this version of PIPO lacks parentheses They were not provided because the additional complexity would not have been sufficiently useful for the 24 development of the artistic operations for which IPMS was intended The last kind of value a PIPO operation can refer to is an externally supplied array value The value calculated so far is used as an index into the array with a value of zero selecting the first element of the array For example to map each pixel of a picture into a new value one simply says DO 0 0 T This simple operation causes the value of pixel 0 0 to be used as an index into the array The value found in the array becomes the new current value An extension of this technique permits the taking of histograms by using the increment operator I DO 0 0 IT The current value is used as an index into the table and the value found there is incremented To set and examine the table the user employs the abbreviation TABLE The value of this abbreviation is a string of numbers representing the values of successive elements of the array For example the commmand SET TABLE 02 4 6 8 10 25 sets the first six elements of the table to the specified numbers The DO command reads the value of TABLE when it begins and updates it when it is done In order to implement some commercial artists operations another PIPO capability was required the ability to execute sequences of operations conditionally based
12. to the right Note also that an operation which must refer to pixels to its left or above itself has to consider a pixel other than 0 0 to be the current pixel The operation will thus end up translating the picture to the left and or up These annoyances are not serious because IPMS provides other commands for translating pictures in any direction PIPO commands can also combine several pixels To produce an output picture summing four adjacent input pixels one simply says 23 DO 0 0 0 1 1 0 1 1 Spaces within a PIPO specification are ignored Each output pixel will be the sum of four input pixels To divide the result by 4 for scaling purposes another operator is added on the end DO 0 0 0 1 1 0 1 1 4 Note that constants are represented by preceeding them with an octothorpe f Note also that PIPO commands are performed from left to right rather than according to algebraic precedence All of the standard arithmetic and logical operations are implemented As another example DO 1 0 0 0 implements a simple vertical edge detector A more complicated command DO 0 0 0 140 2 1 0 1 2 2 0 2 1 2 2 8 1 1 2 1 1 performs contrast enhancement by first calculating the average value of the neighbors of 1 1 then subtracting 1 1 from it and doubling and negating this difference and adding it to the original value This is less than a completely natural expression of the algorithm because
13. 0 1 0 2 2 0 4 3 0 2 4 0 Since PIPO does not handle parentheses it is necessary to say instead DO 1 0 2 S 1 0 2 0 4 S 2 0 3 0 2 1 0 2 0 4 0 Here the store operator S is used to store the products back into the input array elements They are then added back in This operation is slow because of its use of memory to hold partial results and clumsy Furthermore it does not generalize to operations requiring more than one input scan line the optimization described above leaves around the old scaled values Adding parentheses to the language would not be hard and is clearly desirable For efficiency it ought to use whatever registers are available to hold partial results storing into temporary memory locations only when absolutely necessary 39 Another problem with PIPO as implemented is that it can generate only one output pixel per iteration Some operations could be expressed in a more efficient format if they could write out many pixels in one iteration For example the shadow generating algorithm could work by scanning for a diagonal edge that is a configuration in which a black point is diagonally adjacent to a white one At such an edge the algorithm could write out a diagonal line of shadow pixels 1 Another extension which would be useful for picture processing would be the ability to calculate with floating point pixels Although this extension would not be difficult it woul
14. AN INTERACTIVE PICTURE MANIPULATION SYSTEM by Daniel Lewis Franklin SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREES OF BACHELOR OF SCIENCE and MASTER OF SCIENCE at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY September 1978 Signature of Author veec60690602622 46020606029 656 654 e6 2297 909 00 09 269290240 99996 Department of Electrical Engineering and Computer Science August 11 1978 Certified by 06 9 9209 9 0 9 6 6 we e 9 e Vo 3 e o e ojooro 9 60 069 00 95 90 95 85 099 99 9 q Ph Sis Supervisor MIT Certified by ee 9 amp e 6 9 e 9 9 6 9 9 e 9 6 9 9 9 9 9 6 9 9 6 9 amp 6 iens ent 1A _ Company Supervisor Gell vu Accepted EE EE SE E E OJ OG OK NJ OK JO ON ON OK MI Chairman Departmental Committee on Graduate Students ARCHIVES MASSACHUSETTS INST UTE G7 TECHNOL OGY AN Wr WS JAN 31 1543 LIBRARIES An Interactive Picture Manipulation System by Daniel Lewis Franklin Submitted to the Department of Electrical Engineering and Computer Science on August 11 1978 in partial fulfillment of the requirements for the Degrees of Master of Science and Bachelor of Science ABSTRACT A system for manipulating scanned picture files is described The system features an interactive language PIPO for quickly and efficiently performing local position invariant operations There are also commands for inter alia filling in an arb
15. all that is needed is the necessary software Hence one of the tasks of the system under development was to automate as much of the procedure as possible The procedure can be divided into two parts specification of the area to be screened and generation of a file of phototypesetter commands for producing the complete ad As discussed in chapter 2 the artist indicates the area to be screened by drawing curves on the picture surrounding the 43 areas to be filled Then at the console he indicates one point anywhere in each area to be filled The FILL command then fills in the area contained in the curve The filled in area is displayed on the TV Filled in areas can be changed in shade or erased entirely Once the artist is satisfied he writes out the picture A non interactive task will later convert the picture into phototypesetter commands IPMS was used to develop and check out the filling algorithm Its FILL command was then rewritten for the NOVA target machine The FILL command regards all closed curves as convex curves with inner projections stalactites and stalagmites To see how it works consider the following algorithm GIVEN a white picture containing a closed curve in black a point x y within the area enclosed by the curve DO xl x yl y repeat scan left along the line y yl current line until left edge black point scan right along the line until right edge black point doing find
16. ation including the dimensions of a subpicture and the colors which should be used to display each pixel value on the TV Furthermore picture files would not have to be in the scanned bit map picture format which IPMS commands expected they could for example be in some compressed format which the bottom level routines could automatically convert Even without these extensions IPMS proved to be a convenient system on which to develop algorithms for operations on scanned pictures 59 OT UJM pTeTIeAO VI AT bra VI 30 eurTano aT bra p FIN f A LE CREE VEEL DAVOD de 27 DE d VI SO MOPEIS co Dp tbid pagas A y
17. bbreviations were stored was a file of IPMS commands other parameters which were not abbreviation values such as TV characteristics could be set As a bonus a complete demonstration sequence could be prepared in advance The system was not without its problems The elementary picture I O routines lacked any notion of subpictures These were implemented at a much higher level see the description of the DO command This higher level provided two windows one applied on input and one on output As long as the input picture and the output picture did not change the effect was identical to that which would have been obtained by associating the subpicture specification with the individual pictures However any time several different pictures were operated upon or an output picture fed back as input the windows proved remarkably clumsy Although the original specification of the I O routines included subpicture operations for the sake of simplicity they were not implemented It was felt that providing them at the lowest picture manipulation level made operations not requiring them needlessly complicated and inefficient 57 This was a mistake First nearly all practical picture operations require subpictures even though this was only a development system such operations were annoyingly frequent Second the simplicity obtained by leaving out the capability was an illusion the needed complexity was merely pushed up to a higher far mo
18. be by an edge If the center pixel is 0 then it is indeed next to an edge so put out a white point value 1 Otherwise the center pixel must be in the middle of white space so put out a black point value 0 Figure 1b contains an example of this operation 2 Edge generation This operation produces a copy of the input picture with the insides of white areas hollowed out The technique is to simply check whether the current pixel is completely surrounded by white if so it is turned black DO 0 0 amp 0 1 amp 0 2 amp 1 0 amp 1 2 amp 2 0 amp 2 1 amp 2 2 IF gt 0 ELSE 1 1 The ampersand amp is used to indicate logical AND The first part of the specification will have a value of 1 if and only if all the neighbors of 1 1 are 1 white The IF operator has no preceeding C operator so the IF will behave as though a comparison had been made with zero 2T 3 Shadowing This consists of generating a shadow of the picture omitting the picture itself The picture and the shadow will later be combined using two separate colors The algorithm simply produces a white pixel whenever 1 the current pixel is black and 2 a white pixel is within a fixed distance in a diagonal direction The shadow lies along the axis of the diagonal DO 0 0 IF 0 ELSE 1 112 213 314 415 5 This command puts out a black pixel whenever pixel 0 0 is white Otherwise it puts out a white pixel if any of the pixels on the diagonal within
19. bpicture the user could also OR and XOR the two subpictures together This ability turned out to be only marginally useful in the development of the picture processing operations described 31 Chapter III The DO Command Introduction The SEL 86 is a 32 bit single address machine with 8 32 bit general registers Its instructions generally come in two forms immediate in which the operand is a constant contained in the instruction and memory reference A memory reference instruction can specify an index register and or an indirect address pointer chain to be used in address calculation It also specifies the size of the operand byte halfword fullword or doubleword Thus there is add immediate add byte add halfword etc A memory cycle on SEL 86 is 600 nsec Each simple immediate instruction takes one cycle each memory reference instruction takes one cycle longer than its immediate counterpart or longer still if indirect addressing is specified As discussed earlier on page 9 interactive picture operations of the sort envisioned for IPMS set severe constraints on execution efficiency and user convenience rendering both assembly language and FORTRAN unacceptable What was needed was a compile and go translator providing the efficiency of assembly language and the convenience of FORTRAN or better Such a translator is provided by the IPMS DO command 32 Implementation of the DO Command The DO command translates
20. code were to be as efficient as PIPO The programming could be made less tedious by using FORTRAN but at a great cost in execution time Furthermore a complete compile link load cycle would still be required for changes in the algorithm PIPO avoids these problems by providing a compile and go translator The language is simple and heavily influenced by the instruction set of the machine but is still a great improvement over the alternatives Other Picture Manipulation Systems Graphical systems can be divided into two general types those which work in terms of shapes points lines curves etc and those which work with scanned pictures pictures stored as a two dimensional collection of points Historically systems which work with scanned pictures have not offered nearly as many operations as the shape drawing systems nor have they been easy to augment with new operations An early example of a picture generation system is SKETCHPAD 1 which permitted the definition of complex line 1 Sutherland I E SKETCHPAD A Man Machine Graphical Communication System Ph D Diss Dept of Electrical Engineering MIT Cambridge Mass 1963 16 drawings using a light pen and vector CRT display Objects once drawn could be readily replicated Their shapes could be made dependent on constraints permitting for example demonstrations of lever mechanisms However SKETCHPAD was limited to line drawings entered at the CRT
21. d not be useful for the artistic operations for which IPMS was intended and so was not done at this time Additional flexibility could be provided if the user were given the ability to return out of an iteration This would enable one to do pattern searching reasonably fast Of course there would need to be some way of communicating the coordinates of the pattern It might also seem useful to allow the user to call his own special purpose routines This extension however removes one of PIPO s chief benefits its interactiveness The user would have to go through the same tedious process of assembling linking and loading that the DO command was intended to avoid It would seem more worthwhile therefore 1 This shadowing algorithm is due to Don Davis of PRIME Incorporated 40 to increase the power of the language itself so that such a mechanism is unnecessary 41 NAME ge HON I ELSE FI RL SL SLA L LN Notes 1 2 3 THE PIPO OPERATORS OPERAND TYPES 1 FUNCTION memory logical OR memory logical AND memory exclusive OR memory constant add memory constant subtract memory constant multiply memory constant divide memory constant compare for IF 2 none test on condition K gt K gt O NOD3 none alternate branch for IF none terminates IF ELSE clause memory increments the location memory stores into named location RR bits rotates left or right SR bits shi
22. ere necessary in order to be able to process this information at a reasonable rate The complete algorithm is described in chapter 4 Unfortunately the fill command was not always successful when applied to digitized line drawings The curve to be filled in often had small gaps The fill command could escape through these gaps if they were not eliminated To solve this problem a command was added to draw lines at 30 specified places on the picture As with the fill command the line command accepted coordinates which were obtained with the tablet reading command The user was asked to specify a pair of points and a line was drawn between them For convenience the command combination accepted any number of pairs of points until the user indicated that he was done by pressing a switch on the tablet A third command wrote out the picture in a form suitable for conversion to phototypesetter code and another non interactive command did the actual conversion The conversion was not done immediately because it took much time and resulted in a larger set of data than the picture itself These commands are also discussed in chapter 4 IPMS also provided a command level syntax for combining pictures In its simplest form it permitted a user to replace any rectangular section of a picture with a rectangular portion of another or the same picture simply by setting the subpicture windows appropriately In addition to replacement of the su
23. file of commands it is read in by the command file facility just like any other command file Additionally it may be examined by the user and edited at will Thus as much or as little of the IPMS environment as desired may be restored Note that by using the same facility which is nothing more than the ability to store strings by name and recall them at will for many similar purposes IPMS is kept simple This simplicity of design provides a high degree of functionality without requiring excessive memory The commands which manipulate picture files are also kept simple by simplifying the environment in which they do I O IPMS can deal with picture files having from 1 8 bits per pixel up to 2048 pixels per line and any number of lines IPMS commands however deal only with pictures having 8 bits per pixel that is one pixel per addressable unit of Storage They obtain their data by calling the picture I O routines which keep track of the actual size of pixels and pack or unpack them as required The size of a pixel the width of a line and the number of lines must be fed to the picture I O open routine However even this job is centralized the command processor interprets a special argument syntax to indicate the name of a picture file to be opened and looks up the format of the picture file in the abbreviation facility If the format has not been specified the command processor will interrogate the user for it Once 19 it
24. five pixels are white As simple as this algorithm may appear the shadows it produces are quite acceptable See figures 1c and id All the DO commands given so far have assumed that one iteration is done for each and every pixel in the subpicture being used as input Many picture operations require otherwise sampling algorithms for example or ordered dither To permit skipping pixels the DO command looks up the values of the abbreviations DX and DY Their values must be positive integers DX is the horizontal stepsize DY is the vertical stepsize They are both normally one The SET command may be used to change their values 28 The usual use for DX and DY is in displaying digitized pictures on the TV These pictures usually have a much higher resolution than the TV Hence setting DX and DY to a suitable value and executing the command DO 0 0 will display a sampled picture on the TV Some operations required more flexibility than that provided by the DO command In particular in order to implement the screening operation two new commands were needed 1 A command to interactively specify the area to be screened 2 A routine to convert the picture of the screened area into a series of phototypesetter commands The easiest way to specify an area to be screened was through a closed curve to be filled with screening of a specified shade This was because the picture generally already contained curves around areas to
25. fts left or right SRA bits like SL and SR but signed memory constant loads the value memory loads negated value memory is either a matrix point specified as X or X Y for X Y nonnegative integers if Y not given 0 assumed or the T operand constant is of the form lt integer gt where 32768 lt integer lt 32767 bits is a nonnegative integer ranging up to 31 inclusive If there is no C operator immediately before an IF comparison with 0 is assumed Calculations are carried out in 32 bits Overflow can be tested for with the IFO if overflow and IFNO if no overflow operators 42 Chapter IV Screening and the FILL Command Introduction Screening is a process for simulating gray scale using reproduction methods normally incapable of it Gray areas are represented by arrays of small dots The size of the dots determines the shade of gray This technique can be used to represent halftone photographs which contain continuously changing gray levels or just to add patches of gray of various shades to black and white pictures Presently these patches of gray are added by hand a craftsman carefully cuts transparent windows in overlays to match the areas to be screened These overlays are then combined with sheets of screening and photographed The result is combined with the original picture Sophisticated typesetters are readily capable of producing both the black and white and screened portions of the ad
26. h to the L operator in the table which generates a load instruction This is how the first value specified is loaded for use If the first few characters were a constant 34 for example the L operator would be found It generates a load immediate instruction just like the other immediate instructions Before beginning translation and after it is over fixed sequences are put out The prefix sequence starts loop execution the suffix sequence stores the value left in the register all the operators use and branches back to the beginning if the loop has not yet been executed the specified number of times Here is an example The IPMS command DO 0 0 1 is translated into the loop copy reg r0 r5 load im r2 lt number of iterations loop load r1 Xaddr1 5 r3 add im r1 1 store r1 lt addr2 gt r2 add im r3 lt stepsize gt ine br r2 100p return The two instructions preceeding the label loop set up the loop for execution Strictly speaking the copy is unnecessary it is provided as a matter of convenience for the 36 calling routine The load immediate initializes register 2 with the negative two s complement of the number of iterations which is inserted in the instruction before execution The loop is then executed Register 1 holds the current value The store instruction outputs the current value into the appropriate output pixel Then the index register for the input array register 3 is bumped by the step
27. ine for convenience and later digitized using the scanner 2 Performing modifications and additions on those pictures including combining them and screening portions of them 3 Generating special lettering 4 Producing a file of phototypesetter commands which reproduces the completed ads A system meeting these requirements is now under development The system will run on a NOVA 830 16 bit minicomputer with 48K words running mapped RDOS The hardware includes three Diablo moving head disks a 9 track tape drive a graphics tablet a Dest Data digital scanner a 512 x 512 color TV display a Versatek printer plotter and a Tektronix 4010 console To aid in its development another system was needed which would provide an environment suitable for command development The minicomputer itself was not particularly hospitable as it was slow and lacked memory Once a command for e g drop shadowing was developed it could be recoded in NOVA assembler for efficiency but for testing purposes an environment in which one was not penalized very much for inefficiency was highly desirable The problem then was to develop picture processing operations for a commercial page layout system within the framework of a general interactive picture manipulation system The operations to be developed included screening drop shadowing haloing and outlining The Interactive Picture Manipulation System IPMS was the result IPMS is imple
28. ions to selected areas especially since BUGSYS is not interactive Moreover there are no facilities for combining two or more pictures The XAP system 2 developed at the University of Maryland allows logical and arithmetic operations to be performed using several previously digitized pictures but provides no mechanism for introducing or deleting features of a picture It also is not interactive Users must write FORTRAN programs which call the appropriate XAP routines It is thus Similar to a subroutine package The CIPG system is an interactive facility capable of operating on several scanned pictures simultaneously combining them scaling them and performing many other common operations on them New operations however must be defined by writing new commands there is no facility for specifying a 1 Ledley R S et al BUGSYS A Programming System for Picture Processing Not for Debugging Comm ACM Vol 9 p 79 1966 2 Hayes Kenneth C Jr XAP User s Manual Technical Report 348 Computer Science Center University of Maryland 1975 12 picture processing operation and seeing its behavior immediately Commercial artists can manually erase offending portions of a picture crop the picture superimpose parts of other pictures on it draw on it and generally make drastic changes to it The IPMS is intended to possess the same capabilities and to provide them in a more convenient form than the photographer
29. itrary closed curve translating a picture file into a file of phototypesetter commands and compressing a picture file into a compact format suitable for long term storage The system employs an extensible command language which permits the user to easily define his own commands in terms of those already existing Its modular design also aids in the writing of entirely new commands Donald E Troxel Associate Professor Dept of Electrical Engineering ACKNOWLE DGEMENT I wish to thank my supervisor at Bell Labs Dr Peter Denes for providing helpful criticism and a guiding hand Prof Troxel my thesis advisor for his advice Susan Patterson of Bell Labs for all her help both during my stay at Bell and afterwards and Gary Madison and Mark Levine for providing moral and financial support I also wish to thank Jerry Roberts of ECD Corp for allowing me to use ECD s SMART ASCII terminal without which this thesis would probably still have gotten done but not without a great deal more pain Finally I thank the Student Information Processing Board of M I T for funding computer time for the preparation of this thesis This thesis was prepared using the Honeywell Multics Word Processing System CONTENTS Chapter I II III IV Design GOdlS z 9 ee Rip ais eee 9 wel ae See tare A EA CPOG UCC OM ed cece eee wh ero o VVU wie we UR Bie os eel ce B Other Picture Manipulation Systems eee IPMS Struect re
30. leftmost white point on the line below push its coordinates if the point above is white and if the point to its left is black push its coordinates if the point below is white and if the point to its left is black push its coordinates fill in the current line pop the last point pushed assign to xl yl until there are no more points to pop The inner loop can be streamlined somewhat by combining the search for the first white point on the line above or below with the search for points after stalactites or stalagmites Demonstrating this optimization requires somewhat more detail xl x yl y repeat loop for x xl step 1 until point x yl black x left edge x prev point above black prev point below black loop for x x left edge41 step 1 until point x yl black if prev point above black and point x yl 1 white push x yl 1 if prev point below black and point x yl l white push x yl 1 point x yl black prev point above point x yl 1 prev point below point x yl 1 until pop xl yl returns error 47 One further complication ensued It was required that the command take as input the picture containing the curve and produce as output a separate picture containing the filled areas These areas would later be translated into screened regions while the original would be translated into black and white The algorithm above modifies the picture as it goes In fixing this problem i
31. losophy This is the philosophy of providing a small set of general primitives with as little overlapping functionality as possible so that while a single command might not do very much most tasks could be undertaken by combining them in various ways Furthermore if the primitives are not sufficient to perform a given task only those additional capabilities needed to complete the task must be programmed there is no need to re invent the wheel for each new application This permits faster development of new processes by encouraging a building block approach to command development an approach which has been very successful on UNIX 1 In order to encourage it however the command processor should provide ways of coupling individual commands It must be possible to execute a series of commands as though they were one command Each command may need to communicate information such as picture coordinates to other commands Obviously the user should not have to memorize such information and type it back in Ideally the user should not need to be aware of the TS P a P 1 Ritchie D M and Thompson K The UNIX Time Sharing ocu Comm of the ACM Vol 17 No 7 July 1979 pp 365 375 15 communication process at all except when he needs to influence it Yet another requirement for the system was that it handle interfacing with the pec
32. mented using the following equipment 1 An SEL 86 computer 600 nsec memory cycle time with 40K 32 bit words of core hardware floating point and two fixed head disks he cx 2 A raster scanned color television display with 512 X 508 picture elements each up to 4 bits long and stored in a digital refresh memory the red blue and green video signals are generated automatically by accessing the refresh memory 3 A Dest Data 1 2 x 11 digitizing seconds scanner digitizer capable of scanning an 8 black and white picture no gray scale at a density of 240 lines inch in about 5 4 An 11 x 11 tablet and stylus Computek GT50 10 with 1K x 1K resolution 5 An alphanumeric keyboard a Tektronix 611 storage scope and the system 6 A PDP 11 45 high speed system for a Tektronix 4601 Hard Copy Unit used as console running MERT UNIX and attached through channels to the SEL 86 It provides a file the SEL using several 65 megabyte moving head disks and timesharing services for editing programs to be run on the SEL Only the SEL however operates the interactive displays ske IPMS includes a language PIPO for describing a wide class of local position invariant picture operations As defined by Rosenfeld 1 a position invariant picture operation is one in which the effect of the operation on a point does not depend on its position within the picture i e it is a func
33. name of a command it checks to see if the user has a file by that name If so that file is used for further input just as though the user had typed its contents When the file is finished IPMS returns to the console for further input The abbreviation facility is somewhat more complex It permits the user to abbreviate any command or part of a command or series of commands by specifying the desired abbreviation and the command sequence to the SET command The abbreviation can then be used whenever desired by typing a period at the beginning of an argument The rest of the argument will be taken as an abbreviation and its value will be substituted in place of the argument Because so many abbreviations are just commands if the abbreviation is the name of the command to be executed no period is necessary The abbreviation facility is also accessible to commands which use it to save values of coordinates and other information intended to be invisible to most users It is also used to keep certain information about pictures used during the session Much of th s information e g abbreviations set up during a session should be saved for the next session with IPMS This ability is provided by the SAVE command It writes out all the information in the form of a series of SET commands which when executed set up the abbreviations variables etc 18 to have the same value they did when the command was executed Since it is just a
34. nditional branch its earlier matching IF is made to point to the location just after the unconditional branch and the location of the unconditional branch is stacked in place of that of the IF When the end of the specification is reached the stack is checked If it is not empty the specified locations are filled in with the address of the instruction after the last operator This is the behavior that would have occurred anyway if the specification had ended with the appropriate number of FIs This default action was chosen to minimize the number of FIs needed in most operations In practice it is rare to need any FIs at all The T operand also requires special treatment This operand refers not to a constant or a pixel value but to the value of an element in a separately supplied array The element is chosen on the basis of the current value This operand is used for table lookup It can be used with any memref operator It assembles into the same general form as the matrix specification that is an address and an index But the address is the address of the base of the array that is element 0 The index register is the register holding the current value Thus the operator uses the appropriate value from the array If the translator does not recognize the first few characters it prefixes an L to them and tries again If the 35 first few characters specify a coordinate such as 0 0 this action will generate a matc
35. on comparison of the last calculated value with other values Using the conditional feature of PIPO one can compare the value generated so far with a constant or element of the matrix performing some other operation conditionally as a result The C operator is used to perform the comparison and one of the IF operators is used to act on it For example DO 0 0 C 0 1 IF lt 0 ELSE 1 This command compares each pixel with its right hand neighbor for each pixel that was less than its neighbor the output picture will contain a 0 otherwise it will contain a 1 Any number of commands may follow an IF or ELSE and they may refer to the value last generated All the commands following an IF are conditional upon the IF until a succeeding ELSE or FI FI rarely needed closes the IF statement so that later commands are performed regardless of the result of the IF All the normal comparisons are provided IF lt IF IF IF lt IF gt and IF not equal With the operators discussed so far it is possible to implement the following operations 26 1 Outlining This operation consists of drawing lines around the edges of the picture For a black and white picture this is simply DO 0 010 110 211 011 212 012 CHO IF 1 112 2 1 C40 IF 1 ELSE 0 ELSE 0 This operation ORs together all the neighbors of 1 1 The vertical bar i is used to indicate logical ORing If any of them were nonzero the center pixel may
36. on line and capable of being printed on a CRT phototypesetter in a single pass and shipped to the printer In Yellow Pages production there are many examples of picture manipulation carried out by time consuming photographic techniques The directory customer has a wide variety of picture formats at his disposal which are prepared to his specifications by the telephone company s commercial artists These pictures can contain gray areas of several shades The process of adding these areas is referred to as screening Since it involves the use of a screen of dots to simulate gray scale Unusual typography is also available A commercial artist can be requested to do custom lettering or a conventional font can be used with one or more modifications such as outlining drop shadowing adding shadows to the letters to make them look like three dimensional block letters and haloing adding haloes around outlines of the letters Although advertisements containing all these features can theoretically be produced by CRT typesetters the required Software does not exist Hence these operations are currently performed using photography manual processing and pasteups a time consuming process To automate this work requires a system capable of the following 1 Scanning and digitizing pictures drawn by a commercial artist to make the pictures available for computer processing and scaling them to fit The pictures will be drawn off l
37. otal execution time Note that Successive loop executions generally require overlapping lines to be read in In fact once an initial set of lines has been read in each further use of the loop generally requires one new Scan line if DY is one and discards one old line Therefore it ought to be possible to simply read in the new line over the old one and inform the PIPO program of the new mapping between y coordinates and memory addresses To provide this ability the PIPO translator outputs an unfinished program one in which the mapping between y coordinates of matrix elements and memory addresses is left unspecified A separate routine then fills in these addresses It is possible to relink the program with new addresses after each execution of the loop but it is normally well worth the additional space to simply link N copies of the program one for each cyclic permutation of the N scan lines in memory The appropriate copy is then selected each time a new line is read in The result is an N fold decrease in execution time 38 Extensions to the Language Although PIPO as it is now implemented is easily capable of handling problems such as the operations desired by commercial artists many picture processing algorithms require considerable distortion before they can be implemented using PIPO For example consider a smoothing algorithm which scales the values of a group of pixels It would be convenient if one could say DO 0
38. pesetter output The general purpose abbreviation facility also proved extremely useful It enabled single commands which were abbreviations to be at once powerful easy to use and easy to change Other users and other tasks for IPMS might require totally different sets of abbreviations the system permits any degree of abbreviating without penalizing the user in any way for abbreviations he does not need The abbreviation facility was especially helpful with the DO command Once a PIPO specification was worked out a user would have no desire to retype it every time that operation was desired Abbreviating the DO command with that specification permitted building up a library of picture operations As described in chapters two through four the abbreviation facility also provided a convenient way for commands to communicate with other commands as well as system routines It also provided a place for storage of data needed by a single command from one of its uses to the next minimizing the need for retyping All in all it proved to be a very clean accessible facility 56 Providing permanent storage of abbreviations in textual form rather than some internal binary format also had many advantages The user could without being at the SEL 86 console examine the abbreviations developed during the last session generate new abbreviations or delete obsolete ones all with the system text editor Also because the form in which the a
39. r all the points within the curve at a given y coordinate to be filled it is only necessary to push one of them on the stack at some time That every horizontal line will have at least one of its points pushed follows from the fact that two adjacent lines each have at least one point with the same x coordinate Every horizontal line can eventually be reached through adjacency The algorithm fails however on closed curves not satisfying the above property For such curves it is not true that all the points on a horizontal line within the curve are reachable from one pushed point Consider the curve aeggeage ePoe8qQqoe 20050008 000800 As the line containing point S the starting point is scanned point P will be pushed but point Q will not Clearly what is needed is to identify the beginning of these additional horizontal lines and push the coordinates of them as well The beginning of such a line is indicated by the presence of a black point followed by a white one The coordinates of the white point should be pushed The modified algorithm is as follows 46 GIVEN a white picture containing a closed curve in black a point x y within the area enclosed by the curve DO xl x yl yf repeat scan left along the line y yl current line until left edge black point scan right along the line until right edge black point doing find the leftmost white point on the line above push its coordinates find the
40. r to center distance between rules S L D sqrt 2 where D is the screening density in diagonal lines per inch L is the resolution of the typesetter basic units per inch and S is the center to center spacing in the typesetter s basic units For clarity the rest of this discussion will assume a not atypical resolution of 720 units per inch that is a basic unit of one tenth of a point For the Yellow Pages a screening density of about 90 lines to the inch was chosen as the nearest convenient density to the standard 85 lines inch being employed This density yields a spacing of 12 tenth points Given a center to center spacing of 12 tenth points it can be seen that a rule 12 wide and 6 high is the largest that 49 can be accomodated without overlapping For non overlapping rules the gray level can be calculated from L w H S 5 2 where W and H are the width and height of the rule S is the spacing found above and L is the gray level 0 white 1 black Note that this formula assumes that rules are perfectly black Although this is not strictly true the discrepancy can be ignored for now Adjacent rules that is rules on adjacent lines will overlap if both the width and height of the rule are more than 6 tenth points In this case the gray level is L W H W 6 H 6 S S 2 The second term is the amount of overlapping area How many distinct gray levels can be obtained Ass
41. re visible level Third the loss of efficiency would have been insignificant compared to the losses incurred in doing the actual bottom level I O Therefore one enhancement that ought to be made to the system is a subpicture capability at the lowest level that is in the picture I O routines Another decision made early in the development cycle was to avoid a special picture format so as to be able to work with almost any scanned picture that might be obtained The system considered a picture file as merely a stream of pixels with no formatting information whatsoever This view also enabled convenient examination of pictures of unknown format changing the system s notion of the pixel length picture width or picture height only required the use of one or two commands Decoupling the picture formatting information from the picture itself was probably also a mistake Again by keeping the picture file format simple genuinely necessary complexity was pushed up to a far more visible level While the user only had to specify the picture format once he did need to be aware of the fact that the format had to be saved from session to session Furthermore it would be difficult to extend the 58 _ format description to include other parameters It would have been better to convert the picture into a special internal format when the user first specified its properties This would have simplified the storing of a great deal of per picture inform
42. rror by encoding each scan line individually ending the sequence of pairs for each scan line with a special sequence For the phototypesetter translator program the pixel values will all be either 0 or 1 l can represent either a black pixel or a gray pixel 1 An area of gray pixels is translated into a screened area The algorithm used to generate APS command files attempts to put out as few commands as possible It uses a virtual cursor directed by the run length encoded picture file and a 1 This awkward format which requires a separate picture file for each gray level of the picture was necessitated by the constraints on the target system 51 true cursor which simulates the beam on the phototypesetter For a value of zero white the virtual cursor is moved horizontally by the length specified When the end of line character sequence is detected the virtual cursor is moved down to the beginning of the line The true cursor is not moved If the value is one black then it is necessary to generate rules First the commands needed to move the phototypesetter beam represented by the true cursor are produced and the true cursor is superimposed on the virtual cursor Then the commands to produce a rule or sequence of rules if this is a screening file are generated and the true cursor is moved to the position it will occupy after those commands are executed Since space can be saved if the rule being produced is the same si
43. size of the loop This is the size of the step the virtual matrix takes between iterations that is the value of DX see chapter 2 Finally register 2 is incremented and if it is not yet zero control is returned to the location labeled loop This sequence takes 7 cycles per iteration for a total time of 1 1 seconds when run on a 512x508 picture The entire command actually takes over twice as long because of time spent reading and writing the TV buffer If the command were instead run on a disk file it would take even longer Table 1 summarizes the operators in the PIPO language and their possible operands Reducing I O Overhead The description above has oversimplified the translation task somewhat There is no limit to the size of the matrix coordinates A DO command like DO 0 0 1 0 requires only one input sean line while DO 0 041 1 2 2 3 3 4 4 requires five All those scan lines are needed in order to produce one line of output Hence the PIPO translator determines how many input 37 lines are needed so that the DO command can allocate an appropriately sized buffer and read them in before executing the PIPO loop Note that each use of the loop only produces one output line It would appear that if a PIPO program requires five scan lines for each iteration a fresh set of five lines must be read in each time the loop is executed This is inefficient particularly since I O overhead is such a large portion of t
44. t must be remembered that the filled in areas generated in the course of the algorithm are needed to guide it so that it does not attempt to fill in the same area more than once This is not merely an efficiency consideration without the filled in areas the algorithm will for all but the simplest cases loop forever Hence the fill routine was modified slightly it produced an output picture separate from the given input and read from both the input and output pictures Producing Screened Pictures on the Phototypesetter Once the artist is satisfied with a screened ad he writes it out The picture is then converted to phototypesetter commands The actual commands to perform these operations are different on the IPMS and the target machine Hence the following description applied only on IPMS After screening the complete picture consists of several picture files This includes the original picture and one 48 picture for each shade of gray desired Another command writes out these pictures in compressed format Given the compressed files a non interactive task converts them into a series of phototypesetter commands that generate the picture Producing screened pictures requires merely producing a series of dots across the appropriate regions To make the lines less apparent these points are arrayed along diagonal lines at angles of 45 and 135 degrees to the side of the page The following formula yields the horizontal cente
45. tion only of the values of selected pixels not their coordinates Position invariant picture functions are thus analogous to time invariant functions in conventional signal processing A local picture function is one which to the extent that it is a function of input pixel values only depends on those within a finite area The class of local position invariant picture operations is a large one and it includes most useful picture processing operations For example averaging quantizing and edge detection are all local position invariant operations Thus a system which provides the ability to specify almost any such operation should be quite useful The usefulness of PIPO would be limited however were it not for its high efficiency A 512 by 512 TV picture contains approximately a quarter of a million points An operation on all of them will take 0 15 seconds longer for each additional SEL machine cycle They add up rather quickly Equivalent efficiency could be obtained by programming the function in 1 Rosenfeld A Picture Processing by Computer University of Maryland Computer Science Center Report TR 71 June 1968 Contract Nonr 5144 00 p 1 11 9 assembler but such programming would be tedious and anything but interactive requiring a complete assemble link load cycle for each change in the program A complete cycle would also be required in order to adapt the code for a different number of pixels per line if the
46. uliarities of the outside world differing file formats etc Providing clean interfaces to dirty environments permits even the most casually written command to properly handle them This requirement was especially important in view of the many different file formats that IPMS was expected to handle and the peculiarities of its TV display Overview IPMS consists of the components diagrammed in figure 1 Each of its parts will be discussed in more detail 16 COMMAND FILES t i USER COMMANDS i l 1 t 1 ABBREVIATION FACILITY ee ee Vu Shee Eu ara a ee 1 i Li t 1 COMMAND PROCESSOR t I 1 i i yy OTHER COMMANDS 1 t aa E a DO COMMAND FILL COMMAND I SET COMMAND 1 t 1 1 t t I J m PICTURE I O ROUTINES BOTTOM LEVEL I O ROUTINES CONSOLE The command command syntax PICTURE FILES GRAPHIC TABLET J lt Figure 1 IPMS Block Diagram processor implements a common straightforward Each command is specified by giving the name of the command followed by its arguments The name and arguments are separated by spaces Two facilities are provided by the commmand processor to combine elementary commands command files and an abbreviation 17 facility Command files are just what the name implies a file of commands to be executed If IPMS does not recognize the
47. ume that the phototypesetter is capable of putting out rules as small as 1 by 1 However the Yellow Pages can reliably reproduce dots no smaller than 5 mm in diameter or approximately 3 by 3 A selection of rule sizes and the gray levels they produce is given in table 2 For Yellow Pages purposes only three distinct gray levels are actually required so more than enough are available to choose from There is one minor obstacle to this process Producing one dot requires two phototypesetter commands one to move the beam and one to draw the rule For typical screened areas 50 the number of dots drawn can easily reach twenty thousand or more Such a large number of commands would cause screened pictures to require an unacceptable amount of room both on the tape mounted on the phototypesetter and stored on disk in the process of production Hence the sophisticated instruction set of the particular phototypesetter employed an APS 4 was used to set up a loop in one of its buffers which could draw one line of dots Then only three commands per line were required one to step vertically one to set the length of the rule and one to execute the buffer This procedure reduced the number of commands needed to an acceptable level Recall that a run length encoded picture file is a sequence of value length pairs with each pair representing length consecutive pixels of value value A small amount of protection is provided against e
48. y convenient interactive way of specifying local position invariant picture operations Although simple in both design and implementation PIPO proved to be both powerful and efficient Using it drop shadowing outlining and haloing were easily designed and tested It is expected that future operations of the same general type will be just as straightforward to accomplish For serious picture processing however PIPO is not nearly powerful enough Some extensions to the language which would make it more expressive for operations such as contrast enhancement and averaging were discussed in chapter 3 When it was necessary to design an algorithm for an operation filling which was not position invariant and hence fell outside the range of PIPO s capabilities the modular design of IPMS enabled concentrating on the algorithm while spending very little time worrying about interfacing with the system or the user By splitting a given operation into its component parts for example separating the filling operation into 1 obtaining user input 2 filling in the area and 3 55 producing phototypesetter output it was possible to use the same individual routines for other purposes For example the routine to obtain user input from the tablet was used by the line drawing routine Also the fill command could be used in the composition of full color pictures on the TV even though it was actually used only for black and white phototy
49. ze as the last rule the program also keeps track of rule sizes reusing them when possible In this way the size of the resulting data is kept down The program attempts to produce phototypesetter pictures on a scale of one pixel per tenth point For ordinary black and white picture files this presents no problems For Screening files since dots are considerably larger than one tenth point some finessing is required Horizontally partial rules are put out at each end of a line when needed Vertically the program simply samples every sixth line These compromises produce acceptable pictures For testing purposes however it is convenient to be able to convert a picture file to a screened picture such that each scan line produces one 52 line of dots Otherwise it would have been necessary to pad each line with five lines of zeroes For this reason scaling facilities were provided in the conversion program A scale of Six to one causes a picture file to be represented precisely 53 SCREENING DENSITY AS A FUNCTION OF RULE DIMENSIONS 98 LINES INCH WIDTH HEIGHT GRAY LEVEL 3 3 0 125 3 4 8 167 3 5 9 298 3 6 8 250 3 7 9 292 4 4 0 222 4 5 6 278 4 6 9 333 4 7 9 389 5 5 6 347 5 6 0 417 5 7 0 486 6 6 0 580 6 7 0 583 Table 2 54 Chapter V Conclusions The IPMS provided a convenient framework on which to design and test picture processing algorithms for the Yellow Pages project The language PIPO provided a ver
Download Pdf Manuals
Related Search