GB2048537A - Adaptive OCR front-end system - Google Patents

Abstract

A light source 26 illuminates an information area on a document, the reflected light being detected by a sensor 38. A filter circuit equalizes the outputs of the sensor to produce an equalized video signal. The equalized video signal is then registered in a memory to represent the background signature of the information area. The registered background signature is then used as a reference for comparison with subsequent equalized video signals to produce corrected video signals, the registered background signal being updated as necessary in response to sensed white areas of the document. <IMAGE>

Description

SPECIFICATION Adaptive OCR front end system This invention relates to a system for processing video information from a document using a video sensor.

Prior systems for processing data from a document using a video sensor often required tedious adjustment of a light source and a light guide assembly to illuminate an information area on the document.

Coupling efficiency of the light source and light guide was poor, so overrated light sources were used.

However, heat problems from the excessive power consumption resulted. In addition, the sensor response to changing levels of document reflectances declined with increased frequency of video information exposure; complex video processing circuits, individual variations of light sensitive elements in the video sensors, and nonuniform reflectances of various documents and areas thereof went uncompensated.

The applicant's document lighting portion of the video information processing system offers illumination of the information area in a more efficient manner, thereby reducing both the power consumed and heat generated by the light source. A sixth order LC passive filter, simpler than circuits of the prior art, compensates for the sensor's response degradation to changing levels of document reflectances as the frequency of video information exposure is increased. Sensor element and document reflectance variations are given corrective attention by the digital background signature circuit of the invention.

Video information from a document is processed by a system incorporating a light source in close proximity to a filled, circular pattern of a bundle of light transmitting fibers. By arranging the opposite end of the light transmitting fiber bundle in a filled rectangular pattern close to the document, an information area on the document may be illuminated.

The illuminated area is divided by a defined number of scan lines. Each scan line has a corresponding light sensitive element which detects the video information in that line and emits a pulse output proportional to the sensed reflectance. The many light sensitive elements form an array whose pulse train output is equalized and detected by a sixth order LC passive filter. The equalized sensor outputs, representing the background signature of the document's information area, are then registered in a memory and compared to subsequent equalized sensor outputs. The resulting difference is filtered to shift the level but not the shape of the registered background signature when the general reflectance intensity of the information area rises and falls.The value of the filtered, shifted, registered background signature serves as a reference threshold for comparison with subsequent equalized sensor outputs.

Referring to Figure 1, the system for processing video information from a document utilizes three subsystems. The first includes an apparatus for illuminating the document to sense video information: a second is a circuit that equalizes and detects a pulse amplitude modulated sensor output; while a third is a circuit that learns the document's background reflectance signature in the information area to be used as a reference threshold for comparison with subsequent equalized video.

Referring to Figure 2, the information area 20 of a document 21 is divided into two sections, a document margin 22 supplying the background signature of the document and an area containing printed data 24. The document margin 22 is located within .25 inches (6.35 mm) of the document's right edge where no printed data is normally found. The printed data area 24 of the document follows to the left of document margin 22.

To illuminate document margin 22 and data area 24, a point light source 26 (Figure 3) is preferred which emits a light cone 27 of circular projection. This emitted light cone is then directed to document 21 by a flexible bundle 28 of continuous light transmitting fibers 30. Each fiber 30 is made of flexible glass or plastic and has a diameter of approximately 3 mils (25.4 microns). To reduce light transmission losses from light source 26 to fiber bundle 28, fibers 30 are positioned to close proximity to light source 26 and arranged in a filled circular pattern 32 to optimize the capture of the emitted light cone 27. From this circular capture point, fibers 30 are rearranged in a filled rectangular pattern 34 at the light transmitting end 36 of fiber bundle 28 near document 21. The filled rectangular pattern 34 allows defined illumination of the document's information area 20.In addition, the close proximity of fiber bundle 28 to light source 26, filled circular pattern 32 optimally capturing the emitted light cone 27, and the filled rectangular pattern 34 of fiber bundle 28 near document 21 minimize the light transmission and interface inefficiencies to permit the use of lower powered light sources.

The illuminated information (video) is reflected from document 21 and focused on a sensor 38 (RL64EL, manufactured by Reticon Corporation, 910 Benicia Avenue, Sunnyvale, California 94086) by lens 40 (Figure 1). The sensor 38 includes an integrated linear array of sixty-four light sensitive diodes 42 whose resistances vary upon exposure to degrees of light intensity. Corresponding to the sixty-four light sensitive diodes 42 are sixty-four scan lines 44 (Figure 2) which horizontally divide the information area 20 of document 21. The distance between each scan line is 5 mils (127 microns). As the average amount of reflected light in a scan line 44 is sensed by its corresponding light sensitive diode 42, the sensor 38 will produce a proportional output pulse termed the sampled video (Figure 1).In other words, the sensor's pulse output is modulated by the intensity of the sensed reflected light. When each of the sixty-four diodes 42 is sequentially addressed to sense the reflectance of their corresponding scan line 44, the successive sensor outputs produce a pulse train which approximately represents the sensed video information (Figure 4bur. One sequential address of all the diodes consitutes the scan of a vertical slice line 46 (Figure 2) and the first diode in the sensor 38 will not be addressed again to start a second slice line 46,6 mils (152.4 microns) from the previous slice line, until the reflectance in the sixty-fourth scan line has been sensed by the sixty-fourth diode.

Achieving maximum sensitivity and resolution of the sensor 38 (Figure 1 ) to the illuminated information area 20 (Figure 2) is commonly achieved by making the aperture width of each light sensitive diode 42 (Figure 1) equal to their center to center spacing. The sequential addressing of each diode 42 results in the pulse train output described above having a pulse width corresponding to the diode aperture width and a pulse interval corresponding to the diode center to center spacing (Figure 5).These characteristics of the pulse having a duration and interval both equal to X can be mathematically modeled by the fourier transform:

where the sensor pulse train frequency fs equals 1/s, the frequency of the reflectance changes in the information area is f,, and the magnitude of the sensor output in response to f and fr is P(fr). Table 1 indicates that increasing the frequency of the reflectance changes fr decreases the amplitude of the sensor output P(fr).

TABLE 1 fr P(fr) 0 P(O)= .25fas .9 P(O) .50 fs .637 P(O) .75fas .3 P(O) fs 0 Consequently, without compensation, the frequency of the sensor's pulse train, and thereby the speed of information sensing, would be limited to lower values.

To compensate for the decreasing magnitude of the sensor's output, an equalize? filter 48 (Figure 6) was designed to detect the sensor's output and yield the response characterized by: 1 (Table 2).

P(fr) TABLE 2 P(fr) 0 1/P(O) .25fas 1.111/P(0) .50fas 1 .569/P(0) 7sf3 3.33/P(0) fs 00 In addition, unwanted signal harmonics carrying no information from the circuitry were eliminated by a low pass filter 50 (Figure 6) following the equalizer filter. The determination of a minimum cut-off frequency fc to adequately reconstruct the sampled video information was accomplished by using the ideal Nyquist rate of f312, where f3 is the sensor's pulse train frequency (Figure 7). However, due to inherent variations among sensors 38, the Nyquist cut-off frequency fc was chosen within the range

Using this range, a sixth order LC passive filter 52 (Figure 8), made to realize the equalizerfilter-low pass filter objectives, yields a maximum output at fc (Figure 9). Figure 10 illustrates the sequence of sensing, amplifying, and filtering the sampled video. The sensor's output is preamplified to a level where the LC passive filter is operational. The filtered signal is then amplified to become the equalized video which is sent to the digital background signature circuitry described below.

The circuit for learning the document's background signature (Figure 11) is activated by a document presence sensor output 54. In the absence of a document, the document presence sensor output 54 resets to zero a Random Access Memory 56 (RAM) (93419, manufactured by Fairchild Camera & Instrument Corporation, 464 Ellis Street, Mountain View, California, 94042) which stores the digital background signature. The RAM is reset to zero so that the learned background signature of the previous document does not interfere with the learning of the new document's background signature.

When a document 21 is present and document presence sensor output 54 has activated the background signature circuit (Figure 11), the lighting and sampling systems described in the previous sections allow the video information in each scan line 44 (Figure 2) of the information area 20 to be sensed by its corresponding light sensitive diode 42 (Figure 1). The initial equalized video pulses of a new document are fed to comparator 58 (CA3100, manufactured by RCA Solid State, Box 3200, Somerville, New Jersey, 08876) where they are compared to the value of the digital background signature for the corresponding prior sensed scan line 44 stored in RAM 56 and converted to analog by converter 60 (DAC08BC, manufactured by Datel Systems Incorporated, 1020 Turnpike Street, Canton, Massachusetts, 02021).However, the stored background signature values will be zero due to the document presence sensor output 54 resetting RAM 56 when the new document was sensed. Consequently, the first equalized video pulses for the first slice line 46 will have their unaltered values converted to digital at 62 (MC3452, manufactured by Motorola Semiconductor Products, Incorporated, P.O. Box 20924, Phoenix, Arizona, 85036) according to the following scheme.

VID(I): The Ith sample of the video SIG(I): The Ith sample of the signature ISIG(I): The digital representation of SIG(I) DIFF: Video - signature; a continuous analog signal The amplitude of the equalized video (Figure 1) is directly proportional to the reflectance of the document.

If the video range is VminVIDEOVmax, then Vmin represents the black most and Vmax the white most signal levels. During the time that the document presence sensor output 54 (Figure 11) indicates the absence of a document, the value of each ISIG(I) stored in RAM 56, is reset to zero. When the document presence sensor output 54 indicates a document's presence, the digital background signature circuit begins updating each lSlG(l) so that the signature sample SlG(l) approaches the brightest sample VlD(l) in the slice line 46. The approach to the brightest background value is slow enough so that small "hot spots" of high reflectance are not recorded yet fast enough to capture the desired document background signature. The background signature of the document is learned within the space of the document margin 22 (Figure 2).This updating occurs as follows: VmjnsVID(I) - SIG(1)60 ISIG(I) = ISIG(I) L16VID(I) - SIG(I) < L2 ISIG(I) = ISIG(I) +11 L2SVID(I)-SIG(I) < L3 ISIG(I) = ISIG(I) +12 L3sVID(I)-SIG(I) < L4 ISIG(I) = ISIG(I) +13 L4 < VID(I) - SlG(l)Vrn3x ISIG(I) = ISIG(I) +14 where 0 < Ll < L2 < L4 and L is a voltage level 0S11 < 12 < 13 < 14 and I is increment of time Initially, SIG(I) is zero, therefore VID(I) will equal DIFF at the output of comparator 58.That DIFF value will be converted to digital at A/D 62, added in adder 64 (74283, manufactured by Texas Instruments, Incorporated, P.O. Box 5012, Dallas, Texas, 75222) to the prior digital value of the background signature ISIG(I), selected at multiplexer 66 (74158, manufactured by Texas Instruments, Incorporated, P.O. Box 5012, Dallas, Texas, 75222), and transferred for storage to RAM 56.After the sixty-four light sensitive diodes 42 (Figure 1) of the sensor 38 have generated their output pulses which were counted by bit counter 68 (9316, manufactured by Fairchild Camera & Instrument Corporation, 464 Ellis Street, Mountain View, California, 94042) (Figure 11), a start pulse ST' causes slice counter 70 (9316, manufactured by Fairchild Camera & BR< Instrument Corporation, 464 Ellis Street, Mountain View, California, 94042) to address the initial storage bit in RAM 56 for possible updating as the next slice line reflectances are sensed.

After slice line 46 has been stored in RAM 56, the new values of VID(I) are compared to the stored values of SIG(I) according no the equations above.

The algorithm above does not update the signature when the video VID(I) is darker than the signature value SlG(l) for the corresponding scan line 44 and slice line 46 (Figure 2). When this occurs, a carryout is generated at adder 64 so that the previous value of ISIG(I) is chosen at multiplexer 66 and again stored in RAM 56. Consequently, the digital background circuit (Figure 11) only stores in RAM 56 the brightest reflectance in each sensed scan line 44. To track a slowly changing background reflectance intensity in both directions, a global filter 72 is used to shift the level but not the shape of the learned background signature.

The global filter 72 is a low-pass filter with a long time constant in excess of fifty-four scan lines that filters the DIFF signal. The filtered DIFF signal is added to the SIG value in adder 74to obtain the level shifted background signature (compare Figures 13A(b) and 14A(b)) and act as a reference threshold for comparison with subsequent equalized video signals. This comparison is the corrected video signal generated by subtracting the level shifted background signature from the equalized video at comparator 76 (Figures 1 3B and 14B).

By providing the dynamic threshold characteristics of the background signature circuit, variations between sensors and their light sensitive diodes, document reflectances, and aging light sources are compensated.

Claims (12)

1. A system for processing video information from a document comprising: means for illuminating an information area on the document; means for sensing reflectances in the information area; means for equalizing and detecting outputs of the sensing means; means for registering the equalized outputs of the sensing means to represent a background signature of the information area; and means for comparing the registered background signature with subsequent equalized outputs of the sensing means to produce a characteristic output.

2. The system for processing video information from a document claimed in Claim 1, wherein the means for illuminating an information area on the document comprises: a light source; and a bundle of light transmitting fibers having one end in close proximity to the light source and one end in close proximity to the document.

3. The system for processing video information from a document claimed in Claim 2, wherein the light source is a point light source.

4. The system for processing video information from a document claimed in Claim 2, wherein the bundle of light transmitting fibers has the end near the light source arranged in a filled circular pattern to capture the emitted light and the end near the document arranged in a fillled rectangular pattern to illuminate the information area.

5. The system for processing video information from a document claimed in Claim 1, wherein the means for sensing reflectances in the information area comprises: an array of light sensitive elements whose outputs form a pulse train proportional to the sensed reflectances, each light sensitive element being associated with one of a plurality of scan lines which uniformly divide the illuminated information area.

6. The system for processing video information from a document claimed in Claim 1, wherein the means for equalizing and detecting outputs of the sensing means comprises: means for inverting and filtering the outputs of the sensing means.

7. The system for processing video information from a document claimed in Claim 6, wherein the means for inverting and filtering the outputs of the sensing means comprises an electrical circuit realizing the transfer function of

f3 is the sensor pulse train frequency; fr is the frequency of the reflectance changes in the information area; and P(fr) is the magnitude of the sensor output in response to f3 and fr.

8. The system for processing video information from a document claimed in Claim 7, wherein the means for equalizing and detecting outputs of the sensing means comprises a sixth order LC passive filter tuned to a cut-off frequency fc, where

and f3 is the sensor pulse train frequency.

9. The system for processing video information from a document claimed in Claim 1, wherein the means for registering the equalized outputs of the sensing means to represent a background signature of the information area comprises: a memory to store the background signature; and means for removing from the memory the background signature.

10. The system for processing video information from a document claimed in Claim 1, wherein the means for comparing the registered background signature with subsequent equalized outputs of the sensing means to produce a characteristic output comprises: means for taking the difference in signal magnitudes of the registered background signature and subsequent equalized outputs of the sensing means; means for filtering that difference to shift the level but not the shape of the registered background signature when the general reflectance intensity of the information area rises and falls; and means for using the filtered, shifted, registered background signature as a reference threshold for comparison with subsequent equalized outputs of the sensing means.

11. A system for processing video information from a document substantially as hereinbefore described with reference to, and as illustrated in, the accompanying diagrammatic drawings.

12. Any features of novelty, taken singly or in combination, of the system for processing video information from a document as hereinbefore described with reference to the accompanying diagrammatic drawings.