Character Recognition Method and Apparatus

1. Field of the Invention

This invention relates to character recognition and more particularly to the recognition of character images formed in a memory in which the unknown character is first categorized into a category wherein it and one or more possible characters fit, with a final comparison made between each of the possible characters and the known character of the category.

2. Description of the Prior Art

Prior art systems, by and large, have involved "brute force" type of hardware solutions for the recognition of characters. For example one character at a time may be read and deciphered through a programmed logic array (PLA). The PLA scheme is presently used by manufacturers of optical character recognition devices.

The PLA approach is very fast, as required for reading and recognizing magnetic ink characters on checks, but is quite expensive.

Reading bar codes for grocery store check out operations is quite fast, but is also quite expensive.

In this preferred embodiment, an entire National Retail Merchants Association price tag, consisting of approximately 40 characters, must be read within one second. The expensive "brute force" systems mentioned above would be far too expensive and the microprocessor-microcontroller system of this invention provides adequate speed at low cost.

A microprocessor serving as a central processor unit controls a microcontroller for examining the contents of a random access memory which is an image of, in this preferred embodiment, the image received on the face of a CCD imager. The RAM is arranged in columns and rows of bits which are arbitrarily designated as pixels with white pixels being of a "0" binary value and black pixels being of a "1" binary value.

A read-only memory (ROM) contains instructions for controlling the microprocessor. The microprocessor also has a RAM working memory for storage during its computing operations.

The microcontroller is designed to perform certain functions, such as counting pixels, searching for patterns of pixels and loading the image RAM. Since the microcontroller is designed to perform these functions solely, it is much faster at doing so than any known microprocessor. Therefore, the microprocessor initiates operations of the microcontroller through a microcontroller instruction latch. Once instructed, the microcontroller performs its assigned tasks, signalling the microprocessor when the task is finished. For example, the first unknown character on a tag is desired to be located. The microprocessor then instructs the microcontroller to begin searching the image memory by counting pixels in a vertical direction, perhaps half way up the image before going to the next column and counting pixels. When a transition from white to black is encountered, to make sure that an unknown character has been discovered, the pattern search mechanism requires that the black pixels next be counted for a typical count of eight pixels. If four or more of those pixels are black, then it is determined that a character is present.

The next character to the right is then located, and the next and the next, and so on. This determines a base line for one row of characters. If the price tag was skewed, all the characters (and furthermore all of the base lines of each row of characters) are skewed. The industry format for these tags requires that the skew from character to character be plus or minus two degrees so that, for all practical purposes once having determined the skew of one base line, all the base lines can be considered to be skewed by the same angle.

Once having determined the skew angle, the data array on the tag is rotated by the angle to deskew it. A simple row and column shifting operation is satisfactory. If the skew angle is tan.sup.-1 N/M, then the Kth row is shifted over by a number of pixels equal to the integer quotient of KN divided by M, and likewise for the columns. The net result of this operation is a slight magnification of the character along one axis and a demagnification along the other. However, since this effect is less than one unit cell in magnitude normally, the binary quantization of the data suppresses this effect entirely.

A "window" is formed around the first unknown character and a penetration is made from top and bottom and each side of the window to determine the number of white pixels until the black pixels of the unknown character are encountered. The pixels within the window are all weighted and then the white pixels from each side from top and bottom are summed. In addition, the upper and lower half of each side and the left and right half of both top and bottom are separately considered and summed to further categorize the unknown character. Finally, the white pixels defined within the black pixels of the character are counted with all of these summations then being combined. Such combination yields possible characters from the character set. At this point, an actual bit by bit comparison with the pattern character is made for a determination of the unknown character's identity.

A not insignificant part of this character recognition scheme is the removal of dirt. For example, if a single black pixel is found surrounded by white pixels, it is eliminated. Also, after deskewing, bumps on characters are sought out and eliminated. The final result is a character that is imperfect but recognizable.

The major object of this invention is to provide apparatus for rapidly determining the identity of a plurality of characters in an efficient and economical manner.

This major object and other objects will be made evident in the detailed description that follows.

FIGS. 1a and 1b represent a block diagram of the character recognition apparatus of this invention.

FIGS. 2a and 2b represent a schematic diagram of the microprocessor and decoder circuitry of this invention.

FIGS. 3a and 3b schematically illustrates the circuitry for interfacing with the device serviced.

FIGS. 4a and 4b represent a block diagram of the address control shown in FIG. 1.

FIGS. 5a and 5b schematically illustrates the event counter, pixel counter and pattern search circuitry of FIG. 1.

FIGS. 6a and 6b schematically illustrates the instruction latch for the microcontroller and the microcontroller.

FIGS. 7a-7k illustrate the map of the bits of the ROM that controls the microprocessor.

FIGS. 8a-8c illustrate the character set used in this embodiment of the invention.

FIG. 9 is a flowchart of the operation of the circuitry of this invention.

FIG. 10 illustrates a skewed character being deskewed and having its "bumps" removed.

FIG. 11 illustrates the National Retail Merchants Association (NRMA) price tag format in which the unknown characters are found.

FIG. 12 illustrates the penetration designations.

FIG. 13 illustrates the summing of pixels.

FIG. 1 illustrates microprocessor 11 being clocked by clock 12 which also has a divider 13 connected thereto to provide a frequency divided by 6 which is the output "CLK-." Microprocessor 11, in this preferred embodiment, is a Texas Instruments Type 9900 microprocessor which is described in detail in the Texas Instruments Incorporated Manual entitled "Bipolar Microcomputer Components Data Book for Design Engineers" dated January, 1977, beginning at Section 3, Page III. Clock 12 is a Texas Instruments Type SN74LS362 Four-Phase Clock Generator/Driver described beginning at Page 7-460 of Texas Instruments Incorporated Handbook entitled the "TTL Data Book for Design Engineers" Second Edition, Copyright 1976. Divider 13 is a Texas Instruments Type SN74S169 described beginning at Page 7-226 of the TTL Data Book. ROM 15 is connected to microprocessor 11 through data bus 18 and address bus 19. The working RAM 16 is also connected to microprocessor 11 through data bus 18 and address bus 19. Image RAM 17 is connected through buffered address bus 22 and transceivers 20 to the address bus 19 and finally to the microprocessor 11. Image RAM 17 is also connected to microprocessor 11 through data bus 18. Decode 14 decodes microprocessor 11 instructions and is connected to microprocessor through address bus 19.

Latch 30 connects to microcontroller 25 and contains instructions for microcontroller 25 which are sent from the microprocessor 11 through data bus 18, transceiver 21 and buffered address bus 22. Event counter 31 is connected to microprocessor 11 using the same path as indicated above with a word from microprocessor 11 providing the instruction for latch 30 and the count for the event counter 31. Address control 27 is connected to buffered address bus 22 and through bias 26 to microcontroller 25. The address control provides a means of addressing image RAM 17 for loading and retrieving data. In this particular embodiment, the image of the NRMA price tag is formed on a CCD imager which is part of a staring wand assembly. The CCD imager of this preferred embodiment is described and claimed in U.S. patent application, Ser. No. 042,749, entitled "Block Output CCD Imager," filed May 29, 1979 now U.S. Pat. No. 4,251,834 issued Feb. 17, 1981, and assigned to the assignee of this invention. Wand interface 32 communicates wand data available to the buffered address bus 22, trigger signal input, with strobe and buffer signal outputs.

OCR buffer 33 receives the trigger and strobe signals from the wand interface 32, is connected to microprocessor 11 through address bus 19 and receives a "pattern found" signal from microcontroller 25. OCR buffer 33 is connected to TRU buffer 34 which is the buffer that communicates with, in this preferred embodiment, a transaction register unit for outputting to that unit the recognized characters. TRU buffer 34 is also connected to the wand interface 32.

Turning now to FIGS. 2a and 2b, the microprocessor 11 and associated circuitry is shown in more detail. In the description that follows, the inverse of a function is indicated by a dash following the function. For example, the inverse of signal ITRIG is indicated as ITRIG-. Signal ITRIG- is applied to terminal IC3 of microprocessor 11 and comes from terminal D0, selector 330 of TRU buffer 34 shown in FIG. 3. Signal IRTY- from the output of inverter 335 whose input comes from latches 342 (FIG. 3b) of buffer 33 is applied to terminal IC2 of microprocessor 11. Signals ITRIG- and IRTY- provides inputs to OR circuit 45 whose output is inverted through inverter 46 and applied to terminal INTREQ- of microprocessor 11. Clock 12 provides four clock phases into microprocessor 11 through resistors 44. A tank circuit comprised of capacitor 42 and inductor 41 is impressed across terminals T1 and T2 of clock 12 with frequency setting crystal 43 connected across terminals X1 and X2 of clock 12. A 12 MHz signal from clock 12 is transmitted to counter 13 which divides by 6 and provides a 2 MHz output which is inverted through inverter 50 providing the output signal CLK-.

Addresses A0-A14 from microprocessor 11 are carried as signals on conductors that form address bus 19 and data lines D0-D15 from microprocessor 11 form data bus 18.

Signal MEMEN- from the microcontroller latch, specifically from three-state output buffer 308 of FIG. 6 provides an input to terminal MEM- of microprocessor 11 to signal that the microcontroller wants the data and address busses. Signal MEMEN- is also inverted through inverter 48 and applied through resistor R1 to one input of NAND gate 49 and directly from the output of inverter 48 to the other input of NAND gate 49. Filter capacitor C1 is connected from the junction of resistor R1 to ground. The output of NAND gate 49 is signal ME- which is applied as an enable input to terminal G2A of each of decoders 141 and 142. Decoders 141 and 142 in this preferred embodiment, are Texas Instruments Type SN74LS138 3-to-8 line decoders, and decoders 143 and 144 are Texas Instruments Type SN74LS139, both described beginning at Page 7-134 of the TTL Data Book. Select signal UA15 is applied to the input of inverter 52 and also to one input of NAND gate 54. Select signal UBSY is applied to the other input of NAND gate 54 and also to one input of NAND gate 53 whose other input is provided by the output of inverter 52. Output signal CDE from NAND gate 53 provides an enable input to decoder 141, while output EVE from NAND gate 54 provides an enable signal input to decoder 142.

Signal DBIN from microprocessor 11 provides an input to inverter 55 whose output provides one input to AND gate 56. The other input to AND gate 56 is provided by signal BWE- from buffer 47 which is the output from terminal WE- of microprecessor 11. The output of AND gate 56 provides signal WBIN-.

Address lines A0, A1 and A2 provide the C, B and A select inputs, respectively, of each of decoders 141 and 142. Address lines A3, A4, A13 and A14 are connected to select inputs 1B, 1A, 2B and 2A, respectively, of decoders 143. Address lines A2, A3, A10 and A11 are connected to select inputs 1B, 1A, 2B and 2A, respectively, of decoders 144. Outputs Y2-Y5 of decoder 141 provides output signals IREI-, IRE3-, IRE5-, and IRE7-, respectively. Outputs Y2-Y5 of decoder 142 provide signals IRE0-, IRE2-, IRE4- and IRE6-, respectively. Output terminal Y6 of decoder 142 is connected as an enable input to terminal 1G of decoders 143. Terminal 1Y3 of decoders 143 provides signal UCE- which is applied as an enable signal to enable input 2G of decoders 143. Output terminal 1Y0 of decoders 143 provides signal REO-.

Outputs Y0 and Y1 from decoders 142 provide inputs to AND gate 145 whose output is connected to the enable terminal 1G of decoders 144. Outputs 1Y0-1Y2 of decoders 144 provides program decode signals PE0- through PE2-, respectively. Output terminals 2Y0 and 2Y1 of decoders 144 provides the CRU decode signals CRE0- and CRE1-, respectively. The CRU decode is simply a decoding for the transaction register unit output.

FIGS. 3a and 3b illustrate TRU buffer 34 and OCR buffer 33. Selectors 330 and 341 are Texas Instruments Type SN74LS251 data selectors described beginning on page 7-362 of the TTL Data Book. Latch assemblies 331, 337 and 342 are all Texas Instruments Type SN74LS259, described beginning at Page 7-376 of the TTL Data Book. Latches 342, selector 341 and associated circuitry form OCR buffer 33, while selector 330, latches 331 and 337, and assocciated circuitry form TRU buffer 34. Buffered address lines A12, A13, and A14 are connected to the select inputs C, B and A respectively of selector 330. They are also connected to select inputs C, B and A of latches 337. Signal CRE0- provides a strobe input to selector 330 and signal CRE1- from decoder 144 provides one input to OR circuit 332 whose other input is signal CRUCLK from microprocessor 11. The output from OR circuit 332 provides the enable input to latches 331. The TRIG signal from the wand interface is applied as one input to NAND gate 333 whose other input comes from the Q1 output of latches 342. The output of NAND gate 333 provides the D0 input to selector 330 and also provides an input, as indicated earlier, for the microprocessor 11. Other inputs to selector 330 are signals DAR from NAND gates 347 and 348, applied to input terminal D3; signal ONELN from the Q2 output of latches 342 and input terminal D2; IRTY-signal from the output of inverter 335 applied to terminal D1 and finally, signal CNTEQ0-, the carry output signal from the event counter up down counter 331 of FIG. 5. Signal CREO- provides one input to OR circuit 334 whose other input is provided by signal CRUCLK from microprocessor 11 and whose output provides the enable input to terminal G of latches 337.

Latches 331 provide output signals STROBE and PCLK from terminals Q7 and Q6, respectively, which go to the wand interface 32. Output signal DAREN- from terminal Q0 of latches 331 provides one input to AND gate 348 whose other input is provided by the output of NAND gate 347. The output of NAND gate 348 provides one input to NAND gate 347 whose other input is provided by the output of inverter 336. NAND gates 347 and 348 form a latch circuit whose output signal, DAR, is inverted through inverter 346 and provided to the TRU interface 35.

Outputs Q0 through Q7 of latches 337 provide inputs D0 through D7, respectively, of selector 341. Signal MSOCR- from the TRU interface provides a strobe input to selector 341 and provides one input to OR circuit 343 whose other input is provided by the output of inverter 344 which inverts signal OCRUCLK from the TRU interface. Data select inputs C, B and A of selector 341 are provided by signals OADDR12, OADDR13 and OADDR14, respectively, from the TRU interface which also provide the latch select signals to terminals C, B and A, respectively, of latches 342. Output signal OCRUIN is provided by selector 341 to the TRU interface. Latches 342 provide for input of data as represented by signal OCRU to terminal D, the output of OR circuit 343 providing the enable signal to terminal G. A clear signal is provided by signal SYSRST- from buffer 345. Output terminal Q0 provides an input to inverter 336; output pin Q3 provides an input to inverter 335 whose output is signal IRTY-. Terminals Q4 through Q7 provide signals XL1 through XL4, respectively, which are applied to the wand interface. These latter signals are for the operator to determine whether all of the rows have read properly from the wand.

The TRIG signal from the wand indicates that the operator has enabled the trigger on the wand which then calls for loading of the image RAM 17 and the microprocessor 11 instructs the microcontroller 25 accordingly.

Referring now to FIGS. 4a and 4b, the address control 27 and bias 26 are shown. The bias 26 is simply a ROM whos ADA, ADB and ADC inputs are provided by signals bias 2, bias 1 and bias 0 from output terminals Q8, Q7 and Q6, respectively, of microcontroller latch 301 shown in FIG. 6. The bias 0 signal also provides an input on terminals B1 through B4 of adder 276 are provided by outputs from ROM 26 on teminals D1 through D4, respectively, and inputs B1 through B4 of adder 275 are provided by ROM 26 outputs on terminals D5 through D8, respectively. Adders 273 through 276 are Texas Instruments Type SN74LS283, described beginning at page 7-415 of the TTL Data Book. Signal lines BA0 through BA3 from buffered address bus 22 provide the A4 through A1 inputs to adder 273; lines BA4 through BA7 provide inputs to input terminals A4 through A1 of adder 274; lines BA8 through BA11 provide inputs to terminals A4 through A1 of adder 275; lines BA12 through BA15 provide inputs to terminals A4-A1 of adder 276. Adders 273 through 276 are Texas Instruments Type SN74LS283 4-bit binary full adders described beginning at Page 7-415 of the TTL data book. These adders are concatinated by connecting together the C0 terminal of one to the C4 terminal of the next. Flip flop array 271 has input lines D1 through D8 provided by output .SIGMA.1 through .SIGMA.4 of each of adders 274 and 273, respectively. Flip flop array 272 has inputs D1 through D8 supplied by outputs .SIGMA.1 through .SIGMA.4 of each of adders 276 and 275. The outputs BA0 through BA7 and BA8 through BA15 from flip flop arrays 271 and 272, respectively, are output on bus 22.

Signal DBIN from microprocessor 11 and signal UCE1- from decoders 143 provide inputs to OR circuit 278 whose output provides one input to AND gate 61. Signal BWE- from microprocessor 11 provides one input to OR gate 279 whose other input is provided by the signal UCE1-. The output of OR gate 279 provides one input to AND gate 62. Signal UBSY-, from microcontroller 25 provides the other input to AND gate 61 whose output signal ADE- provides the input to terminals OE both of flip flop arrays 271 and 272. Signal CLKD- from the output of NAND gate 261 of microcontroller 25 provides the other input to AND gate 62 whose output signal ADCLK- provides the clock inputs to each of flip flop arrays 271 and 272.

FIGS. 5a and 5b schematically illustrates the event counter 31, the pixel counter 28 and pattern search 29. Event counter 31 is made up of Up-Down Counters 331 and 332 which are Texas Instruments Type SN74LS169, described beginning at Page 7-226 of the TTL Data Book. Lines BA8 through BA11 provide inputs on terminals D through A of counter 331 and lines BA12 through BA15 provide inputs on terminal D through A of counter 332. The carry out terminal of counter 332 provides a carry signal into counter 331 whose carry out output, as indicated earlier, provides an input to the OCR buffer 33, and also provides an input to microcontroller 25, specifically as one input to OR circuit 256 of FIG. 6. Signal UCE0- provides the load signal to counter 331 and counter 332. The clock input for counters 331 and 332 is provided by the output of AND gate 333 whose inputs are signals WED- and CLKD- from microprocessor 25.

Signals PATSEL0 and PATSEL1 from the microprocessor latch 30 provide inputs to exclusive OR gate 336 whose output signal PATSEL is inverted through inverter 337 and also provides one input to NAND gate 292. Buffers 334 or 335, gated by signal BA15- and BA15 respectively, provide one input to OR gate 338 whose other input is provided by the output from inverter 337. The output from buffers 334 and 335 also provide the A and B inputs to serial input shift register 291. Shift register 291 is a Texas Instruments Type SN74LS164 8-bit parallel output serial shift register described beginning at page 7-206 of the TTL Data Book. Shift register 291 stores eight bits representing eight pixels which are examined to determine whether they are black or white. The output of buffers 334 and 335 also provides the other input to NAND gate 292 whose output provides one output to NAND gate 293. The other input to OR gate 339 is provided by signal CLKD- and the output signal PCNTCLK starts one of the counters in counter assembly 281. Counter 281 is a dual 4-bit binary counter from Texas Instruments, Type SN74LS393, described beginning at Page 7-489 of the TTL Data Book. The clear signal for these counters is provided by signal PCLR- from microcontroller 25 which is inverted through inverter 282 and applied to terminals 2C and 1C of counters 281. Outputs 1QD through 1QA from counters 281 provide inputs to terminals 2A1 through 2A4 of line drivers 283. Outputs 2QD through 2QA of counters 281 provide inputs to terminals 1A1 through 1A4, respectively, of line drivers 283. Output terminal 1QD also provides the begin count input on terminal 2A of counters 281. Outputs 1Y1 through 1Y4 from line drivers 283 provides output signals D8 through D11. Outputs 2Y1 through 2Y4 provide output signals D12 through D15. Signals DBIN- and UCE0- provide inputs to OR gate 284 whose output provides the output control for the line drivers 283. Thus, the pixel counter is comprised of counters 281, line drivers 283 and miscellaneous circuitry described.

The pattern search comprises PROM 290 and shift register 291, together with supportive circuitry previously described. The PROM is addressed by terminals ADA through ADH which are connected to output terminals QA through QH of shift register 291. If the appropriate number of black (1) pixels are in shift register 291, then the output D01 of PROM 290 will be a "1." Output D01 is connected to NAND gate 293 which has a third output from exclusive OR gate 336, namely signal PATSEL. The output from NAND gate 293 is signal PATFND which is inverted through inverted 294 providing also signal PATFND-, both of which are inputs to microcontroller 25.

Microcontroller 25, together with instruction latch 30 and associated circuitry is shown in FIGS. 6a and 6b. Flip flop array 301 is the instruction latch 30 for microcontroller 25. Inputs on terminal 1D through 8D are provided by lines BA0 through BA7 of the buffered address lines. Outputs terminal Q1 provides an input to inverter 304 which in turn provides one input to OR gate 306. Output terminals Q3 and Q4 of array 301 provides signals PATSEL0 and PATSEL1 mentioned earlier. Output terminal Q5 provides a signal which serves as one input to AND gate 305 whose other input is provided by signal CLKD-and whose output is signal SCLK-, providing a signal to the wand interface 32.

Output terminals Q6 through Q8 provide bias 0 through bias 2, respectively, for the bias PROM 26, described above.

The output of OR gate 306 is buffered through buffer 307 and provides signal WE-. Buffer 308 provides the output signal MEMEN-, buffers 307 and 308 having their outputs controlled by signal UBSY. Signals WE- and MEMEN- are input to microprocessor 11.

NAND gates 252 and 253 form a latch with signal WED- providing one input to NAND gate 252 and signal URST- providing one input to NAND gate 253 whose respective outputs provide the other inputs. AND gate 251 has signal HOLDA from microprocessor 11 as one input and the output from NAND gate 252 providing the other input. The output of AND gate 251 provides the set input on terminal D of flip flop 254. The Q output of flip flop 254 provides signal UBSY to the D input of flip flop 255. The Q output of flip flop 255 is one input to NAND gate 261. Another input to NAND gate 261 is signal PATFND-. The signal CLK-clocks both of flip flops 254 and 255, is inverted through inverter 260 and provides an input to NAND gate 261 and also an input to NAND gate 259. The output of NAND gate 261 provides signal CLKD-, mentioned earlier. NAND gate 259 has another input provided by the Q output of flip flop 255 and has still another input provided by signal PATFND. The output of NAND gate 259 provides one input to NAND gate 257. OR gate 256 has one input provided by the Q- output of flip flop 255 and the other input provided by signal CNTEQ0-, as mentioned earlier. The output from OR gate 256 provides a third input to NAND gate 257 whose output signal URST is inverted through inverter 258, providing output signal URST-, also mentioned earlier.

FIGS. 7a through 7k are a map of the contents of the program ROM 15 which controls the microprocessor 11. The map is coded hexidecimal and illustrates what each bit in the ROM is, that is, whether it is a "1" or a "0." Thus, the entire operation of the microprocessor 11 is defined by the code shown in these FIGS. 7a through 7k.

Having described the circuitry, the operation shall now be described.

FIGS. 8a through 8c show the character set used by the National Retail Merchants Association for the price tag to be read.

FIG. 9 is a flowchart of the operation of the circuitry described in detail in the preceding section of this specification. Clock 70 indicates a "dirt" clean-up with block 71 illustrating what is done. Under the control of microprocessor 11, microcontroller 25 and its associated circuitry is directed to scan the contents of image memory 17, looking for isolated black pixels as indicated by the arrow in block 71. That is, a transition is noted and the black pixels are counted in shift register 291. If there is only one black pixel, it is changed from "1" to a "0," thus removing extraneous matter.

Next, at block 72, a skew determination is indicated. The industry format for the tags is shown in FIG. 11. The absolute maximum information content is 40 characters displayed in a field measuring 1 1/6" high by 1" wide. The unit cell for each character is 1/6" high by 1/10" wide, yielding a 5:3 aspect ratio.

The choice of the CCD imager array size and pixel aspect ratio is dictated by several considerations. First, it is desirable to have the total pixel count be less than 16,384 so that it will fit into a 16K RAM, namely the image RAM 17. A CCD array of 112 vertical by 144 horizontal pixels, (16128 pixels) with a single cell aspect ratio of 3:2, was selected.

The industry specification for the skew parameter is plus or minus 2.degree. from character to character, which for a 9.times.9 matrix is essentially 0. If the price tag is skewed in the field of view of the imager, all of the characters (and furthermore the base lines of each row of characters) are skewed by the same angle. A skew angle is readily determined by going from one character to the next. A simple row and column shifting operation is all that is required to deskew. That is, if the skew angle is tan.sup.-1 N/M, then the Kth row should be shifted over by a number of cells equal to the integer quotient of KN divided by M, and likewise for the columns. The net result of this operation is a slight magnification of the character along one axis and a demagnification along the other. If, however, this effect is less than one unit cell in magnitude (as is usually the case except for very severe skew angles) then the binary quantization of data supresses this effect entirely. The deskewing is indicated at block 73.

FIG. 10 indicates an "R" having being skewed and deskewed by horizontal and vertical shifts finally having been cleaned up. A recognizable "R" is the result.

Block 74 indicates a "bump" clean-up as shown in block 75 at the arrow. That is, any protrusions on letters are changed from "1s" to "0s."

Next, as indicated at block 76, penetrators are employed to categorize the unknown character. Referring to FIG. 12, the various penetrators are shown with the right and left entire size being indicated as penetrators P3 and P9 with the bottom being indicated P6 and the top indicator P12. Further broken down, the upper and lower penetrators for the right side are penetrators P2 and P4, while the upper and lower penetrators on the left side are penetrators P10 and P8. The right and left penetrators on the upper side are penetrators P1 and P11 and the right and left penetrators on the lower side are penetrators P5 and P7.

FIG. 13 illustrates the weighting of the pixels within a window that contains an unkown character. Note that the weighting is progressively increased from the outside portion to the inner portion of the letter. Where no penetration is made, the sum="0" as shown as the left hand side as P9=0. Otherwise, from the bottom, the penetration for the "R" indicates a total count for P6 as 1+2+3+4+1+2+3+4+1+2+1=24. The hole counted within the black pixels is comprised of eight white pixels. It can therefore be seen that a number or character is readily identifiable.For example, #3 has a huge P9 and a P3, P6 and P12 that are virtually zero. In fact, the simple 16- way branch based on the broad window values P3, P6, P9 and P12 alone narrows the possibilities from 24 to 2 (best case) and 9 (worst case) characters.

Referring back to FIG. 9, the "hole" count indicated at block 77 is the counting of white pixels within the black character pixels. The decision tree at block 78 is the next step and a template confirm, at block 79, that is, confirming that a particular unknown character matches a particular known character within its categorization, follows.

Referring to FIG. 1, when the operator of the wand places the CCD imager against the price tag and pulls the trigger, a TRIG signal is applied to OCR buffer 33 and then to microprocessor 11 indicating that a tag is being read and the contents of the CCD imager should be loaded into RAM 17. This is accomplished by the microprocessor 11 instructing microcontroller 25 to down load RAM 17. Wand data comes into buffered address bus 22, through transceiver 21 into data bus 18 and into RAM 17. This down loading is accomplished very quickly.

Next, the memory contents are subject to the "dirt" clean-up mentioned earlier where a scan is made of the contents of RAM 17 by microcontroller 25 under the direction of microprocessor 11. That is, the memory is scanned until a black pixel is detected. If only one black pixel is detected, it is changed from a "1" to a "0."

The next step is to determine where the first line of characters is located. This is done by scanning the RAM image again. This time, the microprocessor 11 will set the pixel count at 50 so that the lower approximate half of the image is scanned rather than the entire image. That is, starting at address 0 in the lower left hand corner of the image, pixels are counted through address 49. The search is again started at address 112 going up 50 pixels and then restarted at address 224 and so on. When a black transition occurs, the black pixels are counted. If four out of eight pixels are black, it is presumed that a character has been detected. The next character in the line is detected in this same manner until all the characters have been detected. Actually the character is unknown and at this time only the bottoms of the characters have actually been detected, providing the skew angle, if any. Then the deskewing operation is performed,

Next, the actual characters must be entirely detected and identified. This is done by again looking for the character with four out of eight black pixels indicating a character. The unknown character is boxed in by addressing each of the corners, knowing how large a character should be and again using the pattern search pixel counter and event counter as required.

After having boxed in the character in a "window" the penetration technique is started. The categorization of each unknown character in the line is then made in the same manner as described earlier.

Th next line is reached by jumping to a point that is closed to the bottom of the next row of pixels as determined by the standards set for the price tag. The line is detected as below with the same operations being performed. The next two lines of characters are handled in exactly the same way. The character information is sent to RAM 16 for further processing. Microprocessor 11, under program control of ROM 15 then compares the categorized unknown character with the characters that it might be until a comparison is made. Then the contents of RAM 16 are sent to the OCR buffer 33, to the TRU buffer 34 and then on command into the TRU through interface 35.

One skilled in the art may make changes in the logic arrays without departing from the spirit and scope of the invention as defined in the appended claims.