JPH0749926A - Character recognizing device - Google Patents

Abstract

PURPOSE:To provide the character recognizing device with enhanced recognition accuracy by improving character recognition speed, and also, decreasing misrecognition caused by an erroneous noise elimination processing. CONSTITUTION:A character image segmented by a character segmenting means 1 is retained in a character image storage means 2. A character discriminating means 3 executes a discrimination processing with respect to an image in the character image storage means 2, and outputs a result of discrimination and a degree of conviction. A character recognition result output means 4 decides whether the result of discrimination is outputted as a result of recognition or not, from the degree of conviction obtained from the character discriminating means 3, and also, in the case it can not be outputted as the result of recognition, a noise elimination processing is executd by a noise eliminating means 5, it is retained in the character image storage means 2, and with respect to the image to which the noise elimination processng is executed, the discrimination processing executed by the character discriminating means 3.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for recognizing a character by cutting out a character image from an input image.

[0002]

2. Description of the Related Art In a character recognition device, if noise is mixed in an input image, there is a problem that the recognition rate is lowered, such as erroneous recognition due to the influence of the noise. Conventionally, in order to reduce false recognition due to the influence of such noise, for example,
As shown on page 8 of the FY1990 Scientific Research Grants Research Results Report "R & D of high-speed and high-accuracy intelligent character system integrating image-type and logical-type information processing," In a recognition device, a method of uniformly removing noise from a cut-out character image before performing recognition processing has been used. Alternatively, for example, as shown in "Introduction to Character Recognition", Ohmsha Co., Ltd., pp. 15-44, there is a method of using feature extraction and identification counting that are not easily affected by noise.

However, the method of uniformly performing the noise removal processing on the cut out image has a drawback that the character recognition speed as a whole is lowered due to the amount of calculation spent for the noise removal processing. In particular, when the proportion of noise included in the image is small, most of the noise removal processing is wasted. In addition, there is a drawback that a portion of the character image that should be included in the character may be erroneously changed by the noise removal process, resulting in erroneous recognition.

On the other hand, the feature extraction / discrimination function that is not easily affected by noise is complicated in processing, requires a lot of calculation processing, and has a drawback that the character recognition speed decreases. Also in this case, an image that does not contain noise can be recognized with sufficient recognition accuracy even by using a simpler feature extraction and identification function. Therefore, when the ratio of noise included in the image is small, There are many wasteful processes.

[0005]

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and improves the character recognition speed as a whole by performing noise removal processing only when it is necessary. An object of the present invention is to provide a character recognition device in which erroneous recognition due to erroneous noise removal processing is reduced to improve recognition accuracy.

[0006]

According to a first aspect of the present invention, in a character recognition device for recognizing a character from an input image, a character slicing means for slicing a character image from an input image and the character slicing means are provided. A character image storing means for storing the character image obtained by the cutting means; a character identifying means for identifying a character from the character image in the character image storing means and outputting an identification result and a certainty factor; Noise removal means for removing noise from the character image and storing it in the character image storage means, and if the certainty factor is greater than a predetermined value, the identification result is output as a recognition result and otherwise. Is an image in which noise is removed from the character image in the character image storage means by the noise removal means, and the image is stored in the character image storage means. It is characterized in that the image noise was removed with a recognition result output means for outputting as a recognition result and the identification result identified by said character recognition means.

According to the second aspect of the invention, in the character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means; and, if the certainty factor is greater than a predetermined first value, outputs the identification result as a recognition result. In other cases, the noise removal means stores the image from which noise has been removed from the character image in the character image storage means in the character image storage means, and the image from which noise has been removed in the character image storage means is identified as the character. If the certainty factor is larger than the second value determined in advance by the means, the identification result is output as the recognition result, and otherwise, rejected. It is characterized in that it has a recognition result output means for outputting a code indicating a recognition result.

According to a third aspect of the invention, in the character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means; and, if the certainty factor is greater than a predetermined first value, outputs the identification result as a recognition result. In other cases, the image from which noise has been removed from the image in the character image storage means by the noise removal means is stored in the character image storage means, and the image in the character image storage means is identified again by the character identification means. When the certainty factor is larger than the predetermined second value, the identification result is output as the recognition result, and in other cases, the noise removal means again uses the noise. It is characterized in that it has a recognition result output means for causing the identification process by the removal process and the character identification means.

In a fourth aspect of the present invention, in a character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means; and, if the certainty factor is greater than a predetermined first value, outputs the identification result as a recognition result. In other cases, the image from which noise is removed from the image in the character image storage means by the noise removal means is stored in the character image storage means, and the image from which noise is removed in the character image storage means is the character identification means. When the certainty factor is larger than the second value determined in advance or there is no noise to be removed by the noise removing means, the identification result is If output as identification results otherwise is characterized in that it has a recognition result output means to perform the identification processing by the noise removal processing and the character identification means by said noise removing means again.

According to a fifth aspect of the present invention, in a character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means; and, if the certainty factor is greater than a predetermined first value, outputs the identification result as a recognition result. In other cases, the image from which noise is removed from the image in the character image storage means by the noise removal means is stored in the character image storage means, and the image from which noise is removed in the character image storage means is the character identification means. If the certainty factor is less than or equal to a predetermined second value and noise to be removed by the noise removal means exists, the noise removal is performed again. The noise removal processing by the means and the identification processing by the character identification means are performed. In other cases, when the certainty factor is larger than a predetermined third value, the identification result is output as the recognition result, and in other cases. Has a recognition result output means for outputting a code representing a reject as a recognition result.

According to a sixth aspect of the invention, in a character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means, and if the certainty factor is greater than a predetermined value, the identification result is output as a recognition result and other In this case, an image from which noise has been removed from the image in the character image storage means by the noise removal means is stored in the character image storage means, and the image from which noise has been removed in the character image storage means is identified by the character identification means. The process which is repeated until the noise to be removed by the noise removing means is eliminated, and the identification with the highest certainty factor among the plurality of identification results obtained. It is characterized in that it has a recognition result output means for outputting the result as recognition result.

According to a seventh aspect of the present invention, in a character recognition device for recognizing a character cut out from an input image,
Character cutout means to cut out a character image from the input image,
A character image storage means for storing the character image obtained by the character cutting means, a character identification means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and the character image storage Noise removal means for removing noise from the character image in the means and storing it in the character image storage means; and, if the certainty factor is greater than a predetermined first value, outputs the identification result as a recognition result. In other cases, the image from which noise is removed from the image in the character image storage means by the noise removal means is stored in the character image storage means, and the image from which noise is removed in the character image storage means is the character identification means. The noise removal means repeats the process for identifying the noise until there is no more noise to be removed, and the maximum value of the plurality of certainty factors obtained is determined in advance. A recognition result output means for outputting the identification result having the largest certainty value as the recognition result when the value is larger than the determined second value, and outputting the code representing the reject as the recognition result otherwise. It is what

According to an eighth aspect of the present invention, in the character recognition apparatus according to any one of the first to seventh aspects, the noise removing means is configured to detect a character image stored in the character image storage means. Noise candidate detecting means for detecting noise candidates, evaluation value calculating means for calculating an evaluation value for each noise candidate detected by the noise candidate detecting means, which is estimated to be noise, and the noise candidate When two or more noise candidates are detected by the detection means, the noise candidate detection means performs noise removal processing from the character image only on the noise candidate having the largest evaluation value of each noise candidate by the evaluation value calculation means. By only 1
When one noise candidate is detected, noise removal processing is performed on the detected noise candidate from the character image, and when no noise candidate is detected by the noise candidate detection means, noise removal processing is not performed It is characterized by having means.

[0014]

According to the present invention, a character image is cut out from an input image, and the character is identified without performing noise removal processing for the first time. When identifying a character, the character identifying means outputs a certainty factor together with the identification result. When the certainty factor is larger than a predetermined value, the identification result is output as it is as the recognition result. As a result, when there is no noise or there is little noise, character recognition is performed without performing noise removal processing, and the character recognition speed can be improved.

Further, the certainty factor by the first character identification is
When the value is equal to or smaller than a predetermined value, noise removal processing is performed and the character is identified again to prevent erroneous recognition due to the influence of noise in the first processing.

In the second aspect of the present invention, when the certainty factor due to the second character identification is equal to or less than a predetermined value, it is determined that the character could not be identified even if the noise removal is performed, and a code representing the reject. By outputting as the recognition result, it is possible to reduce the output of the erroneous recognition result.

In the third aspect, when the certainty factor by the second character identification is less than or equal to a predetermined value, noise removal and character identification are repeatedly performed until the certainty factor exceeds a predetermined value. As a result, the recognition result can be made more reliable.

According to the present invention, it is further determined whether or not there is noise to be removed, and when there is no more noise to be removed, the identification result at that time is output as a recognition result, so that noise removal can be performed infinitely. It prevents repeated identification of characters.

According to a fifth aspect of the present invention, when the identification result to be output is determined, and the certainty factor is less than or equal to a predetermined value, a code indicating reject is determined as a character that cannot be recognized. By outputting the recognition result, it is possible to reduce the output of incorrect recognition result.

According to a sixth aspect of the present invention, when the certainty factor due to the first character identification is less than or equal to a predetermined value, the noise removal process and the character identification process are repeated until there is no noise to be removed, during which time. By outputting the identification result with the highest certainty factor as the recognition result, it is possible to reduce the occurrence of erroneous recognition due to erroneous noise removal processing during repeated noise removal processing. Can be improved.

Further, in the seventh aspect, when the maximum certainty factor is equal to or less than a predetermined value, it is determined that the character cannot be recognized, and the code representing the reject is output as the recognition result, thereby making an error. The output of the recognition result can be reduced.

The above-mentioned noise removing means can be configured, for example, to detect noise candidates from the character image and remove the detected noise candidates one by one from the one having the largest evaluation value. , The noise can be removed in stages. This makes it possible to obtain an appropriate character image in the process of repeatedly removing noise with the character recognition device described in each of the above claims.

[0023]

1 is a block diagram showing an embodiment of a character recognition device of the present invention. In the figure, 1 is a character cutting means,
Reference numeral 2 is a character image storage means, 3 is a character identification means, 4 is a character recognition result output means, and 5 is a noise removal means. The character cutting-out means 1 selects from the input image including a plurality of characters,
An area containing one character is cut out to obtain a character image.
The character image storage means 2 stores the image obtained by the character cutout means 1 or the noise removal means 5. The character identification means 3 performs identification processing on the image obtained from the character image storage means 2 and outputs an identification result and a certainty factor which are composed of a code of a candidate character or a code representing a reject. A known method can be used as the identification processing method. For example, a pattern matching method or a recognition method using a feature extraction method may be adopted. The character recognition result output means 4 determines from the certainty factor obtained from the character recognition means 3 whether or not to output the recognition result as a recognition result, and when the recognition result cannot be output, operates the noise removal means 5. In addition to performing the noise removal processing, the character identification means 3 performs the identification processing on the image subjected to the noise removal processing. This operation is repeated in some cases. The noise removing unit 5 detects and removes noise from the image obtained from the character image storage unit 2, and outputs the image from which the noise is removed. Depending on the processing, when there is no noise to be removed, the fact can be output to the character recognition result output means 4.

The certainty factor output by the character identifying means 3 represents the degree of matching between the image input to the character identifying means 3 and the standard sample of characters. That is, the greater the matching degree, the higher the certainty value. The character with the highest confidence value is output as the identification result. As the standard sample used at this time, a character image containing no noise can be used. Therefore, when the image input to the character identifying unit 3 does not include noise, the degree of matching between the input image and the standard sample of the correct character is large, and the certainty factor output by the character identifying unit 3 is high. The value of becomes large. On the contrary, when noise is included, the degree of matching between the input image and the standard sample of each character is small, and the value of the certainty factor is small. Therefore, it is possible to predetermine the threshold value and estimate whether or not the image contains noise depending on whether or not the threshold value is larger than the threshold value.

FIG. 2 is a block diagram showing an example of the character identifying means 3. In the figure, 11 is an image normalizing means, 12
Is a feature extracting means, 13 is a standard feature storing means, and 14 is a similarity calculating means. The image normalization means 11 detects a circumscribed rectangle that includes all black pixels in the input image,
So that the circumscribed rectangle has a predetermined size,
The normalized image obtained by enlarging or reducing in the vertical and horizontal directions is output. The feature extraction unit 12 extracts a coarse mesh feature from the normalized image obtained by the image normalization unit 11 and outputs it. The coarse mesh feature is obtained by dividing the image into mesh-shaped rectangular regions and extracting the features of the image included in the rectangular region for each of the divided rectangular regions. Known features can be used as the features extracted from each rectangular area. For example, the sum of pixel values is calculated,
This can be characterized. At this time, the coarse mesh features are obtained as a matrix of sums of pixel values. The standard feature storage means 13 stores standard features prepared in advance for each character to be recognized, and outputs the standard feature corresponding to each character in response to a request. The standard features are obtained by normalizing at least one standard sample image prepared in advance for each character by a process equivalent to the process by the image normalizing means 11, and then performing a process equivalent to the process by the feature extracting means 12. The extracted features can be averaged and obtained. The similarity calculation unit 14 calculates the similarity between the feature obtained by the feature extraction unit 12 and the standard feature of each character obtained from the standard feature storage unit 13, and the similarity having the largest value is obtained. The character code corresponding to the standard feature and the maximum similarity value as the certainty factor are output.

FIG. 3 is a block diagram showing an example of the noise removing means 5. In the figure, 21 is an isolated black pixel detection means, and 22 is an isolated black pixel deletion means. The isolated black pixel detection means 21 detects a black pixel in which all the adjacent pixels are white pixels in the image, and outputs the position in the image. If no black pixels satisfying the conditions are detected,
Outputs a code indicating undetected. Isolated black pixel deleting means 2
2 uses the position of the isolated black pixel in the image obtained by the isolated black pixel detection means 21 to replace the corresponding pixel with a white pixel. If the isolated black pixel detecting means 21 obtains a code indicating that it has not been detected, nothing is done.

In one example of this noise removing means, the size of the black pixel detected by the isolated black pixel detecting means 21 can be changed according to the number of times of operation of the noise removing means. For example, the first operation can detect a pixel in which only one pixel is isolated and black, and the second operation can detect a pixel in which two consecutive pixels are isolated. Of course, the black pixel group having a predetermined size can be gradually reduced.

FIG. 4 is a block diagram showing another example of the noise removing means 5. In the figure, 31 is a noise candidate detecting means, 32 is an evaluation value calculating means, and 33 is a noise candidate removing means. In this example, the noise removing means 5 performs the noise removing processing only on the noise candidate having the largest evaluation value estimated to be noise in the image obtained from the character image storing means 2. When the noise removing means 5 operates repeatedly, the removal processing is performed in order from the noise candidate having the largest evaluation value estimated to be noise.

The noise candidate detecting means 31 detects noise candidates satisfying a predetermined condition from the image. The evaluation value calculation unit 32 calculates, for each noise candidate detected by the noise candidate detection unit 31, an evaluation value estimated that the noise candidate is noise. The noise candidate removing unit 33 performs a process of removing the noise candidate from the image for one of the noise candidates detected by the noise candidate detecting unit 31.

In this example, the noise candidate detecting means 31
As a method of detecting noise candidates in, for example, a method of detecting all black pixel clusters in which black pixels are continuous in the image and having an area smaller than a predetermined value as noise candidates can be used. . Further, the evaluation value calculation means 32 calculates the evaluation value by, for example, obtaining the reciprocal of the area of the noise candidate and the distance from the noise candidate to the closest black pixel block after multiplying each by a constant. You can Noise candidate deletion means 33
As the noise candidate to be removed in step 1, for example, the noise candidate having the largest or smallest evaluation value calculated by the evaluation value calculation means 32 can be selected. All the black pixels forming the black pixel block corresponding to the selected noise candidate are replaced with white pixels. The noise candidate detection method, the evaluation value calculation method, and the noise candidate removal method described above are not limited to these methods, and other methods can be used. There is no noise, and the noise candidate detection means 31
When no noise candidate is detected in, the noise removal process is not performed and the input image is output as it is.

FIG. 5 is an explanatory diagram of the noise removing operation in the example of the noise removing means shown in FIG. In the figure, 41 is an image before noise removal, 42 and 43 are noises, and 44 is an image after noise removal. As shown in FIG. 5A, noise 42 and noise 43 exist in the image 41 before noise removal. The noise 43 is the noise generated because a part of the character line is cut off. The noise candidate detection means 31 detects noise 42 and noise 43 as noise candidates. The evaluation values of the two detected noise candidates are calculated by the evaluation value calculation means 32. An example of the calculated evaluation value is shown in FIG. For example, if noise candidates with a large evaluation value are selected, noise 4
The noise candidate corresponding to 2 is selected, and the noise 42 is removed by the noise candidate deletion means 33. An example of an image from which the noise 42 has been removed is shown in FIG. in this way,
It is possible to remove noises that are least related to the character elements in order.

Further, as in the character image shown in FIG. 5C, when the noise 43 is included, for example, in the extraction and identification of the coarse mesh feature in the character identification means 3 shown in FIG. The evaluation value is large, and good results can be obtained. That is, if the noise is erased before the recognition process as in the conventional art, a part of the character such as the noise 43 is also erased, which causes erroneous recognition. Since the character recognition device performs the character identification process before performing the noise removal process, the character recognition device shown in FIG.
Even in the case where a part of the character is erased as noise, it can be accurately recognized.

3 to 4, examples of the character identifying means 3 and the noise removing means 5 are shown, but the present invention is not limited to the configurations shown in these examples, and various known methods may be used. it can.

Some operation examples of the embodiment of the character recognition apparatus of the present invention will be described below with reference to FIGS. 1 and 6 to 12. FIG. 6 is a flow chart for explaining a first operation example in one embodiment of the character recognition device of the present invention. First, in step S51, the character cutout unit 1 cuts out a character image for each character from the input image, and the obtained character image is stored in the character image storage unit 2. In step S52, the character image stored in the character image storage unit 2 is subjected to an identification process by the character identification unit 3. The identification result and the certainty factor are character recognition result output means 4
Is output to. In S53, the certainty factor of the identification result,
Compared with a predetermined value T1, the confidence factor is the value T1.
If it is greater than 1, it is estimated that there is no noise in the image, the identification result is output as the recognition result in S56, and the recognition process for the cut out character ends.
In S53, when the certainty factor of the identification result is not larger than the predetermined value T1, it is estimated that there is noise in the image, and a series of processes including the noise removing process by the noise removing unit 5 is performed. In the first operation example, in S54, the noise removal processing is performed on the image stored in the character image storage means 2 by the noise removal means 5, and the obtained image is stored in the character image storage means 2. In step S55, the character identification means 3 performs an identification process on the image, and the resulting identification result is output as a recognition result in step S56, and the recognition process for the cut out character ends.

FIG. 7 is a flow chart for explaining a second operation example in one embodiment of the character recognition device of the present invention. The processing when the certainty factor is larger than the predetermined value T1 in S51, S52, and S53 is the same as that of the above-described first operation example, and thus the description thereof is omitted. S53
If the certainty factor of the identification result is not larger than the predetermined value T1, the noise is estimated to be present and is stored in the character image storage means 2 in S54, as in the first operation example. Noise removal means 5 performs noise removal processing on the image, and the obtained image is stored in the character image storage means 2. In step S55, the character identifying means 3 performs identification processing on the image. The certainty factor obtained as a result is compared with a predetermined value T3 in S57, and if the certainty factor is larger than the value T3, the identification result obtained by the identification process in S55 is output as a recognition result in S56. Then, the recognition process for the cut-out character ends. Further, in S57, if the certainty factor is not larger than the value T3, it is determined that the noise cannot be recognized even if the noise removal processing is performed.
In S58, the code indicating the reject is output as the recognition result, and the recognition process for the cut out character is completed.

FIG. 8 is a flow chart for explaining a third operation example in one embodiment of the character recognition device of the present invention. The processing when the certainty factor is larger than the predetermined value T1 in S51, S52, and S53 is the same as that of the above-described first operation example, and thus the description thereof is omitted. S53
If the certainty factor of the identification result is not larger than the predetermined value T1, the noise is estimated to be present and is stored in the character image storage means 2 in S54, as in the first operation example. Noise removal means 5 performs noise removal processing on the image, and the obtained image is stored in the character image storage means 2. In step S55, the character identifying means 3 performs identification processing on the image. The certainty factor obtained as a result is compared with a predetermined value T2 in S59, and when the certainty factor is not larger than the value T2, the process returns to S54 to perform the noise removal processing and S55.
The character identification process in is repeated. Then, when it is determined in S59 that the certainty factor is larger than the predetermined value T2, the repetitive process ends. The identification result obtained at that time is used as the recognition result,
The recognition processing for the extracted character is output in S56, and the processing ends.

FIG. 9 is a flow chart for explaining the fourth operation example in the embodiment of the character recognition apparatus of the present invention. This operation example is almost the same as the above-mentioned third operation example, but in S59, when the certainty factor obtained as a result of the identification processing in S55 is not larger than the predetermined value T2, S60 is further added. In, it is determined whether or not there is noise to be removed. Therefore, in the noise removal processing in S54 and the repeated processing of the character identification processing in S55, the certainty factor becomes larger than a predetermined value T2, or the image stored in the character image storage unit 2 is removed. It ends when there is no more noise to be used. By the determination in S60, it is possible to prevent the noise removal processing and the character identification processing from being endlessly repeated due to the inability to be identified. The identification result obtained when the iterative processing is completed is used as the recognition result, and S56
And the recognition processing for the cut-out character ends.

FIG. 10 is a flow chart for explaining a fifth operation example in one embodiment of the character recognition device of the present invention. This operation example is almost the same as the above-described fourth operation example, except that the certainty factor and a predetermined value T are determined at the time when the repeated processing of the noise removal processing and the character identification processing is completed.
3 is being compared. That is, in S59, the certainty factor obtained as a result of the identification processing in S55 is
If it is larger than the predetermined value T2, or if it is determined in S60 that there is no noise to be removed, the certainty factor at that time is compared with the predetermined value T3 in S61. If the certainty factor is larger than the value T3 as a result of the comparison, the identification result is output as the recognition result in S56, and the certainty factor is the value T3.
If it is not larger than 3, it is determined as unrecognizable and S62.
At, the code representing the reject is output as the recognition result, and the recognition process for the cut-out character ends.

In the fifth operation example, when the predetermined value T2 is larger than the value T3, it is obvious that when the certainty factor is larger than the value T2, it is larger than the value T3. When is larger than the value T2, it is also possible to operate so as to output the recognition result in S56 without performing the determination in S61.

FIG. 11 is a flow chart for explaining a sixth operation example in the embodiment of the character recognition apparatus of the present invention. The processing when the certainty factor is larger than the predetermined value T1 in S51, S52, and S53 is the same as that of the above-described first operation example, and thus the description thereof is omitted. S53
When the certainty factor of the identification result is not larger than the predetermined value T1, the noise removal processing is performed by the noise removal means 5 on the image stored in the character image storage means 2 in S54, and the obtained value is obtained. The image to be stored is stored in the character image storage means 2, and in S55, the processing of performing the identification processing by the character identification means 3 on the image is repeated until there is no noise to be removed. At this time, the identification result and the certainty factor output from the character identification means 3 are stored for each of the character identification processing performed in S52 and the character identification processing performed in S55. When it is determined in S60 that there is no noise to be removed, the iterative process is terminated, and S
At 63, of the plurality of identification results obtained by the iterative processing, the identification result with the highest certainty value is output as the recognition result, and the recognition processing for the cut-out character ends.

FIG. 12 is a flow chart for explaining the seventh operation example in the embodiment of the character recognition apparatus of the present invention. This operation example is almost the same as the above-described sixth operation example, but after the repetition processing of the noise removal processing in S54 and the character identification processing in S55 is completed, S64 is performed.
In the above, among the plurality of certainty factors obtained by the character identification processing of S52 and S55, the one having the largest value is compared with the predetermined value T3, and if the maximum certainty factor is larger than the value T3, , S63, the identification result corresponding to the certainty factor is output as the recognition result. When the maximum certainty factor is not larger than the value T3, in S65, a code indicating reject as unrecognizable is output as a recognition result, and the recognition process for the cut out character is ended.

In each of the above-described operation examples, the predetermined values T1, T2, T3 may be determined in consideration of the correctness rate of the recognition result, all of them may be different values, or two or three. It may be set to the same value.

A specific operation in one embodiment of the character recognition device of the present invention will be described. Here, for simplification, the first operation example among the above-described operation examples will be described using a specific example, but the same can be considered for other operation examples. Also, the character identification means 3
As an example, a case where the configuration shown in FIG. 2 is used will be described.

FIG. 13 is an explanatory diagram of an example of a character image.
The character cut-out means cuts out each character from the input image and saves it in the character image storage means as a character image as shown in FIG. FIG. 13A shows a character image without noise, and FIG. 13B shows a character image with noise. In the following explanation, this 2
The processing for one character image will be described.

FIG. 14 is an explanatory diagram of an example of a normalized image. In the character identification means 3, in the image normalization means 11,
Normalization processing is performed on the character image. As described above, the normalization process detects a circumscribing rectangle that includes all black pixels in the character image, and enlarges or reduces the circumscribing rectangle in each of the vertical and horizontal directions so that the circumscribed rectangle has a predetermined size. Then, the obtained normalized image is output. In this example, the normalization processing is performed so that the shape becomes almost square. A normalized image obtained by normalizing the noise-free character image shown in FIG. 13A is shown in FIG. 14A, and a normalized image obtained by normalizing the noise-containing character image shown in FIG. 13B is shown. 14 (B). If there is noise in the character image, the size of the circumscribed rectangle may be regulated by the noise, and in that case, normalization cannot be performed correctly, as shown in FIG.

FIG. 15 is an explanatory diagram of an example of the coarse mesh feature obtained by the feature extracting means. Feature extraction means 12
At, coarse mesh features are extracted from the normalized image. In FIG. 15, the normalized image is divided into 4 × 4 regions,
The number of pixels included in each area is used as a characteristic of each area, and the 16 mesh feature values obtained from each area constitute a coarse mesh characteristic. FIG. 15A is the same as FIG.
It shows the coarse mesh features extracted from the noiseless normalized image shown in (A). In addition, FIG.
It shows a rough mesh feature extracted from the normalized image containing noise shown in FIG. Thus, the two coarse mesh features are different with many different parts.

FIG. 16 is an explanatory diagram of an example of standard features stored in the standard feature storage means. The standard features stored in the standard feature storage means 13 are also created in the same manner as the process shown in FIGS. Here, the process of creating the standard features is shown for numbers, but the same applies to other characters and symbols. FIG. 16 (A) is a standard character image without noise prepared to obtain standard features. An example of a normalized image obtained by normalizing each character image is shown in FIG. Further, FIG. 16 shows an example of coarse mesh features extracted from each normalized image.
It is shown in (C). The rough mesh features shown in FIG. 16C are stored in the standard feature storage means 13 as standard features.

FIG. 17 is an explanatory diagram of the result of character identification by the calculation of the similarity and the output of the certainty factor. The similarity between the coarse mesh features shown in FIG. 15 and the standard features shown in FIG. 16 is calculated by the similarity calculation means 14. Among the calculated similarities, the one having the largest value is set as the certainty factor, the corresponding character code is obtained, and this is set as the identification result. FIG. 17A is the same as FIG.
It is the result of the similarity calculation when the noise shown in (A) is not included. The maximum similarity is 0.99 of “2”. A code indicating "2" is output as the identification result, and 0.99 is output as the certainty factor. In addition, FIG. 17B shows the result of the similarity calculation when the noise shown in FIG. 13B is included. The maximum similarity is
It is 0.76 of "5". A code indicating "5" is output as the identification result, and 0.76 is output as the certainty factor.

In this way, when noise is not included in the image as shown in FIG. 13A, the correct character "2" and a large value are obtained as shown in FIG. 17A. A certainty factor of 0.99 is obtained. However, when noise is included in the image as shown in FIG.
As shown in (B), an erroneous character "5" and a certainty factor of 0.76 with a small value are obtained. Therefore, for example, a threshold value of 0.95 is set in advance, and when the certainty factor is larger than the threshold value of 0.95, it is estimated that the noise is not included. In other cases, the noise is included. It can be estimated that When the certainty factor is greater than 0.95, the identification result is output as the recognition result of the character recognition device. In other cases, the noise in the image in the character image storage means is removed by the noise removing means 5, and the character identifying means again identifies the image.

FIG. 18 is an explanatory diagram of a process of recognizing a character image after the noise removing process is performed. The noise removal processing is performed by the noise removal unit 5 on the character image including noise shown in FIG. As a result, FIG.
As shown in (A), a character image containing no noise is stored in the character image storage unit 2. Hereinafter, the identification process is performed by the same process as the above-mentioned process. First, FIG.
The character normalization image shown in FIG. 18A is subjected to the normalization processing by the image normalization means 11, and the image shown in FIG.
The normalized image shown in (B) is obtained. From the normalized image, the feature extracting means 12 extracts the coarse mesh features shown in FIG. 18 (C). The extracted coarse mesh features,
The similarity calculation means 14 calculates the similarity with the standard feature shown in FIG. The calculation result is shown in FIG. Here, the maximum similarity is 0.99, and this value is used as the certainty factor, and the corresponding character code “2” is output as the identification result. Since this certainty factor is greater than the predetermined threshold value 0.95, "2" is output as the recognition result of the character recognition device. As described above, since the noise is removed by the noise removing unit 5, the correct recognition result “2” can be obtained.

[0051]

As is apparent from the above description, according to the present invention, when the image contains no noise, the recognition result can be output as the recognition result without performing the noise removal processing. Since it is possible to omit the noise processing which is wasteful in the conventional technique, the character recognition speed can be improved.

Further, even if the portion of the image that should be included in the character is erroneously determined as noise, by performing the identification process without performing the noise removal process, the error that has conventionally occurred is eliminated. The erroneous recognition due to the noise removal processing can be almost eliminated, and the character recognition accuracy can be improved.

Further, by providing a noise removing means for removing noise candidates having a large evaluation value estimated to be noise in order, unnecessary noise removing processing is further reduced and erroneous noise removing processing is further reduced. Therefore, the character recognition speed and the character recognition accuracy are further improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of a character recognition device of the present invention.

FIG. 2 is a block diagram showing an example of a character identifying means 3.

FIG. 3 is a block diagram showing an example of a noise removing unit 5.

FIG. 4 is a block diagram showing another example of the noise removing means 5.

5 is an explanatory diagram of a noise removing operation in the example of the noise removing unit shown in FIG.

FIG. 6 is a flowchart illustrating a first operation example in one embodiment of the character recognition device of the present invention.

FIG. 7 is a flowchart illustrating a second operation example in the embodiment of the character recognition device of the present invention.

FIG. 8 is a flowchart illustrating a third operation example of the embodiment of the character recognition device of the present invention.

FIG. 9 is a flowchart illustrating a fourth operation example of the embodiment of the character recognition device of the present invention.

FIG. 10 is a flowchart illustrating a fifth operation example of the embodiment of the character recognition device of the present invention.

FIG. 11 is a flowchart illustrating a sixth operation example of the embodiment of the character recognition device of the present invention.

FIG. 12 is a flowchart illustrating a seventh operation example of the embodiment of the character recognition device of the present invention.

FIG. 13 is an explanatory diagram of an example of a character image.

FIG. 14 is an explanatory diagram of an example of a normalized image.

FIG. 15 is an explanatory diagram of an example of coarse mesh features obtained by the feature extraction means.

FIG. 16 is an explanatory diagram of an example of standard features stored in a standard feature storage unit.

FIG. 17 is an explanatory diagram of an output of a character identification result and a certainty factor by calculation of similarity.

FIG. 18 is an explanatory diagram of a process of recognizing a character image after performing noise removal processing.

[Explanation of symbols]

1 character cutout means, 2 character image storage means, 3 character identification means, 4 character recognition result output means, 5 noise removal means, 11 image normalization means, 12 feature extraction means,
13 standard feature storage means, 14 similarity calculation means, 21
Isolated black pixel detection means, 22 isolated black pixel deletion means, 3
1 noise candidate detecting means, 32 evaluation value calculating means, 33
Noise candidate removal means.

Claims (8)

[Claims] 1. A character recognition device for recognizing a character from an input image, the character slicing means for slicing a character image from the input image, a character image storing means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is larger than a predetermined value, the identification result is output as a recognition result, and in other cases, the noise removing means removes noise from the character image in the character image storage means. The stored image is stored in the character image storage means and the noise-free image in the character image storage means is identified by the character identification means. A character recognition device having a recognition result output means for outputting the identification result of 1. as a recognition result. 2. A character recognition device for recognizing a character from an input image, the character slicing means for slicing a character image from the input image, the character image storing means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is greater than a predetermined first value, the identification result is output as a recognition result, and in other cases, the noise removing means extracts the character image from the character image in the character image storage means. A noise-removed image is stored in the character image storage means, and the noise-removed image in the character image storage means is identified by the character identification means. If the certainty factor is larger than a predetermined second value, the identification result is output as a recognition result, and in other cases, the recognition result output means outputs a code representing the reject as the recognition result. Characterized character recognition device. 3. A character recognition device for recognizing a character from an input image, a character slicing means for slicing a character image from an input image, a character image storage means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is larger than a predetermined first value, the identification result is output as a recognition result, and in other cases, the noise removing means outputs noise from the image in the character image storage means. The image from which the characters have been removed is stored in the character image storage means, and the image in the character image storage means is again identified by the character identification means to determine the certainty factor beforehand. If it is larger than the determined second value, the recognition result is output as a recognition result, and in the other cases, the recognition result output is made to perform the noise removal processing by the noise removal means and the recognition processing by the character identification means again. A character recognition device having means. 4. A character recognition device for recognizing a character from an input image, the character slicing means for slicing a character image from the input image, the character image storage means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is larger than a predetermined first value, the identification result is output as a recognition result, and in other cases, the noise removing means outputs noise from the image in the character image storage means. The image from which the noise is removed is stored in the character image storage means, and the image in the character image storage means from which noise is removed is identified by the character identification means. If the reliability is larger than a second predetermined value or there is no noise to be removed by the noise removing means, the identification result is output as a recognition result, and in other cases, the noise removing means again A character recognition device comprising a recognition result output means for performing a noise removal processing and an identification processing by the character identification means. 5. A character recognition device for recognizing a character from an input image, the character slicing means for slicing a character image from the input image, the character image storing means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is larger than a predetermined first value, the identification result is output as a recognition result, and in other cases, the noise removing means outputs noise from the image in the character image storage means. The image from which the noise is removed is stored in the character image storage means, and the image in the character image storage means from which noise is removed is identified by the character identification means. If the reliability is less than or equal to the predetermined second value and there is noise to be removed by the noise removal means, the noise removal processing by the noise removal means and the identification processing by the character identification means are performed again. Otherwise, the certainty factor is the third
A character recognition device having a recognition result output means for outputting the identification result as a recognition result when the value is larger than the value of, and otherwise outputting a code representing a reject as the recognition result. 6. A character recognition device for recognizing a character from an input image, the character slicing means for slicing a character image from the input image, the character image storage means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is greater than a predetermined value, the identification result is output as a recognition result, and in other cases, the noise removing means removes noise from the image in the character image storage means. A process of storing an image in the character image storage means and identifying an image in the character image storage means from which noise is removed by the character identification means. A character recognition device comprising a recognition result output means for outputting, as a recognition result, an identification result having the highest certainty among a plurality of identification results obtained by repeating the noise removal means until there is no noise to be removed. . 7. A character recognizing device for recognizing a character from an input image, a character slicing means for slicing a character image from an input image, a character image storage means for storing the character image obtained by the character slicing means, Character identifying means for identifying a character from the character image in the character image storage means and outputting an identification result and a certainty factor, and noise for removing noise from the character image in the character image storage means and storing it in the character image storage means Removing means, and if the certainty factor is larger than a predetermined first value, the identification result is output as a recognition result, and in other cases, the noise removing means outputs noise from the image in the character image storage means. The image from which the noise is removed is stored in the character image storage means, and the image from which the noise in the character image storage means is removed is identified by the character identification means. The process is repeated until there is no more noise to be removed by the noise removing means, and when the maximum value of the plurality of certainty factors obtained is larger than the second predetermined value, the identification result having the highest certainty value is recognized. A character recognition device comprising a recognition result output means for outputting as a result and otherwise outputting a code representing a reject as a recognition result. 8. The noise removing means detects a noise candidate from a character image stored in the character image storage means, and a noise candidate detecting means for each noise candidate detected by the noise candidate detecting means. Evaluation value calculation means for calculating an evaluation value estimated to be noise, and when the noise candidate detection means detects two or more noise candidates, the evaluation value calculation means evaluates each noise candidate. Noise removal processing is performed on the character image only for the noise candidate having the largest value, and when only one noise candidate is detected by the noise candidate detection means, noise is detected from the character image for the detected noise candidate. A noise candidate deleting unit is provided which does not perform noise removing process when no noise candidate is detected by the noise candidate detecting unit. The character recognition device according to claim 1, wherein the character recognition device is a character recognition device.