Character Recognizing Method and Apparatus Thereof

The present invention relates to character recognizing method and apparatus and, more particularly, to character recognizing method and apparatus capable of detecting a segmentation point of a character from a character line, thereby recognizing an individual character.

In a conventional apparatus of this type, a method is proposed in which upon recognition of a character line, an average pitch of one character is calculated and characters are segmented and recognized one by one from the character line in accordance with information about the average pitch.

In the above conventional character recognizing apparatus, however, each character is recognized on the basis of the average pitch. Therefore, if characters in a character line to be recognized are arranged at non-uniform pitches, detection is ended in the middle of a character. As a result, characters cannot be read or are erroneously read.

It is, therefore, an object of the present invention to provide character recognizing method and apparatus capable of recognizing characters one by one even if the characters are arranged at non-uniform pitches in a character line.

In order to achieve the above object of the present invention, there is provided a character recognizing method of segmenting characters one by one from a character line and recognizing the characters, comprising steps of detecting a character using an end of the character line as a start point, normalizing the detected character, comparing the normalized value with a predetermined standard pattern, and detecting a point having a minimum distance difference from the standard pattern as a segmentation point between neighboring characters.

In addition, in order to achieve the above object of the present invention, there is provided a character recognizing apparatus for segmenting characters one by one from a character line and recognizing the characters, comprising a detecting means for detecting a character using an end of the character line as a start point, a normalizing means for normalizing the detected character, a comparing means for comparing the normalized value with a predetermined standard pattern, and a segmentation point detecting means for detecting a point having a minimum distance difference from the standard pattern as a segmentation point between neighboring characters.

FIG. 1 is a block diagram showing an embodiment of the present invention;

FIGS. 2A and 2B, are graphs showing operation characteristics of the apparatus; and

FIG. 3 is an enlarged view showing a character line in FIG. 2A.

Prior to a description of an embodiment of the present invention, an operational principle will be described first. When a character line is to be detected as a video signal, this video signal can be considered such that its main and sub scan directions change along the time base. As the video signal gradually changes, a specific character or character line is two-dimensionally formed. In this case, a video signal of a character line having non-uniform pitches is offset or expanded/compressed along the time base direction of the sub scan direction as compared with a video signal of a character line having a uniform pitch. In order to recognize a character line having a non-uniform pitch, therefore, a difference and expansion/compression along the time base must be corrected.

A description will be made with reference to graphs in FIGS. 2A and 2B showing operation characteristics. Assume that a character line (TyPe) in which neighboring characters are in contact with each other at non-uniform pitches as shown in FIG. 2A is an input video signal, and that the input signal is given as follows:

The vector a.sub.i represents the feature of an ith input video signal in a sub scan direction and corresponds to an input video signal of one line in a main scan direction. The input video signal A is a time series of the feature vectors and is a set of input video signals of lines 1 to I. A standard pattern (dictionary) B(n) of n character lines is given as follows: ##EQU1## The vector b.sub.j is a jth feature vector in a sub scan direction of the standard pattern. For the convenience of description, assume that the number of lines of a single standard graphic pattern is 20 and J=20n (FIG. 2A shows a case wherein n=4). S.sub.n (.alpha.) represents a standard pattern, and a symbol .alpha. represents one of a character categories 0 to 9, a to Z and A to Z. Assume that segmentation points of characters in the input video signal A are i.sub.1, i.sub.3, i.sub.5 and i.sub.7 and a function obtained by projecting the segmentation points on the standard pattern is given as follows:

This function is called a distortion function. If the distortion function is known, the sub scan direction of the input video signal can be converted by this function to adjust a time base i along the sub scan direction of the standard pattern B(n). In the case of a character line in which neighboring characters are in contact with each other at non-uniform pitches, however, the distortion function cannot be calculated, and therefore another means is required. That is, if a distortion is present between the input video signal and the standard pattern, the similarity between the two is reduced, and a difference on the time base i is increased. On the contrary, if the input video signal is distorted along the sub scan direction to be most similar to the standard pattern, the distortion on the time base i is eliminated, and an optimal distortion function is determined. Therefore, assuming that the points i.sub.1, i.sub.3 and i.sub.5 are located at the points i.sub.2, i.sub.4 and i.sub.6, respectively, in FIG. 2A, the difference between the input video signal and the standard pattern on the time base i is minimized. For example, as shown in an enlarged view of FIG. 3 showing the character line in FIG. 2A, on the basis of a line P not exceeding the width of one character, a character is detected on lines P+1, P+2, P+3, P+4, . . . The detected signal is then normalized and compared or collated with the standard pattern S.sub.1 (.alpha.), thereby detecting the line P+3 (the point i.sub.1 in FIG. 2B) on which the difference of the time base i, i.e., a noncoincidence count is minimum. The line P+4 is then considered as a start line to repeat the above processing. By performing such processing for all the characters, a line on which the difference of the time base i is minimum can be detected. Referring to FIG. 2B, the points i.sub.1, i.sub.3, i.sub.5 and i.sub.7 represent the detected lines, and the standard patterns S.sub.1 (1), S.sub.2 (3), S.sub.3 (5) and S.sub.4 (7) corresponding to these values are character recognition results of respective characters.

As described above, if the distortion function is unknown, a character on lines is detected on the basis of the line P not exceeding the width of one character, and the detected signal is normalized and subjected to the character recognition processing. In this case, on the basis of the position of each line at which a difference between the normalized value and the standard pattern, i.e., the noncoincidence count is minimum, a segmentation position of each character is detected.

Upon character recognition, although the video signals on lines P, P+1, P+2, P+3 and P+4 may be time-sequentially processed, they may be simultaneously processed to increase the operation speed of the character recognition processing. Alternatively, if the processing is started at the same time from both the first line (the first end of a character line) and the last line I (the last end of the character line) along the sub scan direction, the processing speed can be further improved.

An embodiment of the present invention will be described below. FIG. 1 is a block diagram showing an embodiment of the present invention. Referring to FIG. 1, a detecting means comprises an input video signal storage memory 10, a normalizing means comprises character height detectors 31 to 34 and normalizing circuits 41 to 44, a segmentation point detecting means comprises character collating circuits 51 to 54, a standard pattern unit 60, a minimum noncoincidence count detector 70 and a character determination result storage memory 80. In addition, a control unit 20 is included as a recognizing means.

The input video signal storage memory 10 stores an input video signal as shown in FIG. 2A output from a character detecting unit (not shown). The memory 10 outputs the input video signal to a line l2 in accordance with an address signal output from the control unit 20 to a line l1. On the basis of a control signal output from the control unit 20 to a corresponding one of lines l3 to l6, each of the character height detectors 31 to 34 temporarily stores the input video signal supplied from the memory 10 via the line l2 and detects the height of the character. The control signal supplied to each of the lines l3 to l6 gives the width of the input video signal in the sub scan direction. The line l3 corresponds to the input video signal from the origin to the line P shown in FIG. 2A; the line l4, that from the origin to the line P+1; the line l5, that from the origin to the line P+2; and the line l6, that from the origin to the line P+3.

The normalizing circuits 41 to 44 convert the input video signals into normalized patterns on the basis of input video signals with different widths output from lines l7 to l10 and character height information output from lines l11 to l14, respectively. The patterns normalized by the normalizing circuits 41 to 44 are output to the character collating circuits 51 to 54 via lines l15 to l18, respectively. The character collating circuits 51 to 54 compare and collate the normalized patterns with the standard pattern from the standard pattern unit 60 and output noncoincidence counts in the sub scan direction together with categories to the minimum noncoincidence count detector 70 via lines l20 to l23, respectively. The minimum noncoincidence count detector 70 detects the minimum number in the noncoincidence counts from the character collating circuits 51 to 54 and outputs a line number corresponding to the minimum value to the control unit 20 via a line l24. For example, in FIG. 3, this line number corresponds to the line P+3. The circuit 70 outputs a category corresponding to the minimum noncoincidence count to the character determination result storage memory 80 via a line l25. On the basis of the line number information from the line l24, the control unit 20 outputs a signal obtained by incrementing the previous address signal by one to the input video signal storage memory 10 via the line l1, thereby performing recognition of one character. By repeatedly executing the above series of processing, the entire character line can be recognized.

As described above, according to the present invention, even if neighboring characters are in contact with each other at non-uniform pitches in a character line, each character can be recognized by obtaining a segmentation point between the characters.

In the above embodiment, printed characters are used as an input video signal. According to the present invention, however, hand-written characters or graphic patterns can be recognized.

In the above embodiment, a series of processing to the line P+3 are simultaneously performed. The number of lines of parallel processing, however, is not limited but can be arbitrarily set.

As has been described above, according to the present invention, input video signals of a character line are normalized and then compared with a predetermined standard pattern to detect a segmentation point of each character, thereby recognizing each character. Therefore, even if characters are arranged at non-uniform pitches in a character line, each character can be effectively recognized.