JP2012230482A - Evaluation image processing apparatus and evaluation image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an evaluation image processing apparatus and an evaluation image processing method capable of simply performing a character recognition type performance evaluation by using a uniquely digitized character recognition type evaluation index.SOLUTION: A character recognition processing part 11 performs character recognition for an evaluation image inputted from an evaluation image input part 10 in accordance with a character recognition method specified by an operator and calculates a certainty factor of a recognition result in each character. An average certainty factor calculation part 12 calculates an evaluation value of the evaluation image based on the calculated certainty factors. Although an average value of the certainty factors is calculated as the evaluation value of the evaluation image, calculation of a total value of certainty factors e.g. can be also adopted. An evaluation index classification part 13 classifies the evaluation image based on the calculated evaluation value. An evaluation index output part 14 displays the classified evaluation image and its classification level.

Description

  The present invention relates to an evaluation image processing apparatus and an evaluation image processing method for grasping and evaluating the performance of a character recognition method.

  Until now, character recognition has been performed by a method of reading a document taken from a scanner by a computer or a method of reading a package or the like by a dedicated industrial product in a food manufacturing process, for example.

  Recently, with the spread of camera-equipped mobile phones, there is an increasing need to recognize characters from images captured from cameras. In this case, not only when characters are written on a blank sheet as in a document, but also characters may be written on a scene photo or landscape photo, and techniques for recognizing characters from these are also being considered. .

  When recognizing characters in a scene or landscape photo, the accuracy tends to be lower than conventional character recognition methods such as reading a document imported from a scanner on a computer, and the performance of the character recognition method is evaluated. Thus, it is important to improve the performance of the character recognition method based on the evaluation result. As a performance evaluation method of the character recognition method, the method of Non-Patent Document 1 below is known.

ICDAR 2003 robust readingcompetitions: entries, results, and future directions (IJDAR (2005) 7: 105-122)

  In the technique of Non-Patent Document 1, the complexity of a test set of characters for evaluating performance is divided into three and analyzed. These three are text position recognition, character recognition, and word recognition. However, when this method is used, since the complexity evaluation index is divided into three, the method itself has a problem that it is difficult to handle.

  Accordingly, an object of the present invention is to provide an evaluation image processing apparatus and an evaluation image processing method capable of easily performing performance evaluation of a character recognition method using a uniquely quantified evaluation index of the character recognition method. To do.

  Therefore, in order to solve the above-described problem, an evaluation image processing apparatus of the present invention is an evaluation image processing apparatus that performs evaluation of a character recognition method, an input unit that inputs an evaluation image, and the input that is input by the input unit. Character recognition processing is performed on the evaluation image, and the evaluation value of the evaluation image is calculated based on the character recognition processing means for calculating the certainty for the recognition result for each character and the certainty calculated by the character recognition processing means. Evaluation value calculating means.

  The evaluation image processing method of the present invention is an evaluation image processing method of an evaluation image processing apparatus that evaluates a character recognition method. An input step of inputting an evaluation image, and the evaluation image input by the input step A character recognition processing step that performs character recognition and calculates a certainty factor for the recognition result for each character, and an evaluation value calculation that calculates an evaluation value of the evaluation image based on the certainty factor calculated by the character recognition processing step And steps.

  According to this invention, character recognition is performed on the input evaluation image, a certainty factor for the recognition result is calculated for each character, an evaluation value of the evaluation image is calculated based on the calculated certainty factor, The evaluation images are classified based on the calculated evaluation value. Thereby, it becomes possible to evaluate using a unique numerical index, and the evaluation image can be evaluated by a simple process.

  In the evaluation image processing apparatus of the present invention, the evaluation value calculation means calculates an average value of the certainty factor per character from the certainty factors of all the characters in the evaluation image, which is calculated by the character recognition processing means. May be calculated as

  According to this invention, the evaluation image can be properly evaluated by calculating the average value of the certainty factor as the evaluation value.

  In the evaluation image processing apparatus of the present invention, the evaluation value calculation means uses a total value of certainty factors of all or a predetermined number of characters in the evaluation image calculated by the character recognition processing means as an evaluation value. You may make it calculate.

  According to this invention, the evaluation image can be evaluated by calculating the total value of the certainty factors as the evaluation value. In this case, when the number of evaluation objects is about the same, an appropriate evaluation can be made in comparison with other evaluation images.

  The evaluation image processing apparatus of the present invention further includes an evaluation image classification unit that classifies the evaluation image based on the evaluation value calculated by the evaluation value calculation unit.

  According to the present invention, the evaluation images can be classified based on the evaluation value. When this classification is made plural, it can be evaluated from various angles. For example, a certain character recognition method is suitable for recognizing an image including a large number of characters, and another character recognition method can perform high-speed recognition processing on an image including a small number of characters. For example, it is possible to grasp the performance of the character recognition method and the recognition characteristics. Therefore, it is possible to easily grasp the function-enhanced portion necessary for performance improvement.

  According to the present invention, evaluation can be performed using a unique numerical index, and an evaluation image can be evaluated using a simpler evaluation index.

It is a block diagram which shows the function structure of the evaluation image processing apparatus 100 for evaluating a character recognition system. 2 is a hardware configuration diagram of an evaluation image processing apparatus 100. FIG. It is a schematic diagram which compares the template image and evaluation image in performing a character recognition process. It is explanatory drawing which shows the correspondence table of a level and average reliability. 3 is a flowchart showing processing of an evaluation image processing apparatus 100. It is explanatory drawing which shows the specific example of the evaluation image evaluated and classified by the evaluation image processing apparatus 100, and its certainty factor.

  Embodiments of the present invention will be described with reference to the accompanying drawings. Where possible, the same parts are denoted by the same reference numerals, and redundant description is omitted.

  An evaluation image processing apparatus 100 that evaluates and classifies a character recognition method according to an embodiment of the present invention calculates an average certainty factor that serves as an evaluation index from an input evaluation image, and classifies the evaluation image based on the average certainty factor It is.

  FIG. 1 is a block diagram showing a functional configuration of an evaluation image processing apparatus 100 for evaluating a character recognition method. As shown in FIG. 1, the evaluation image processing apparatus 1 includes an evaluation image input unit 10 (input unit), a character recognition processing unit 11 (character recognition processing unit), an average certainty calculation unit 12 (evaluation value calculation unit), an evaluation An image classification unit 13 (evaluation image classification means) and an evaluation index output unit 14 are provided.

  FIG. 2 is a hardware configuration diagram of the evaluation image processing apparatus 100. As shown in FIG. 2, the evaluation image processing apparatus 100 shown in FIG. 1 physically includes a CPU 1, a RAM 2 and a ROM 3 as main storage devices, an input device 4 such as a keyboard and a mouse as input devices, a display, and the like. Output device 5, a communication module 6 that is a data transmission / reception device such as a network card, an auxiliary storage device 7 such as a semiconductor memory, and the like. Each function described in FIG. 1 has the input device 4, the output device 5, and the communication module 6 under the control of the CPU 1 by reading predetermined computer software on the hardware such as the CPU 1 and the RAM 2 shown in FIG. This is realized by reading and writing data in the RAM 2 and the auxiliary storage device 7. Hereinafter, each functional block will be described based on the functional blocks shown in FIG.

  The evaluation image input unit 10 is a part for inputting an evaluation image that is a test image for evaluating the character recognition method. For example, it is a scanner or a camera, and is a means for capturing a scene photo or landscape photo as an evaluation image.

  The character recognition processing unit 11 performs character recognition processing based on a character recognition method designated by an operator from a plurality of character recognition methods (character recognition engines) set for the evaluation image input by the evaluation image input unit 10. This is a part for calculating the certainty factor for the recognition result. The character recognition processing unit 11 determines one character recognition method based on an instruction from the operator, and performs character recognition processing according to the character recognition method. The character recognition processing unit 11 may be configured to select a single character recognition method to be evaluated and set a plurality of character recognition methods, or to evaluate a character recognition method each time the character recognition method is evaluated. A method may be set and stored.

  The certainty obtained by performing the character recognition processing according to the designated character recognition method is an index representing the certainty of the character data obtained from the result of the character recognition processing, and is used as a template when performing character recognition, for example. It can be obtained by comparing the template image and the evaluation image.

FIG. 3 is a schematic diagram for comparing a template image and an evaluation image when performing character recognition processing. First, with respect to the character region M extracted in the character recognition process (FIG. 3A), the binarization process of the image is performed. “0”. Next, the character area is divided into a plurality of blocks (FIG. 3B). Thereafter, a process of adding the pixel values in each divided block is performed, and the sum value for the template image is Soriginal and the sum value for the evaluation image is Snoisy (FIG. 3C). In the example of FIG. 3C, Soriginal = 25 and Snoizy = 12. A difference between the calculated Soriginal and Snoisy is calculated, a mean square error (MSE) based on the difference is calculated, and it can be calculated as a certainty factor (see Expression (1)). N in the formula (1) is the number of divided blocks. The certainty factor is determined on the basis of the sum of the mean square errors for each of the divided regions to be evaluated. Note that the mean square error may be referred to as an average error or an average square error.

Further, as shown in Equation (2), a root mean squared error (RMSE) may be calculated as the certainty factor. The root mean square error may be referred to as a mean square square error, a mean square error, a mean square error square root, a square root mean square error, a standard deviation of errors, and the like.

  The average certainty calculation unit 12 is a part that calculates an average certainty that is an average value per character from the certainty calculated by the character recognition processing unit 11. That is, the average certainty calculation unit 12 counts the number of recognitions of characters recognized as characters or characters recognized by the character recognition processing unit 11 and recognizes the total value of all the calculated certainty factors. The average confidence is calculated by dividing by a number. This average certainty is a numerical value indicating how much type characters may be included in the evaluation image. The complexity of the evaluation image can be determined using this numerical value.

  Note that instead of the average certainty factor calculation unit 12, the evaluation value calculation unit may calculate all the characters that are the recognition targets in the evaluation image or the total value of the top N certainty factors as the evaluation value. . When evaluation images with the same number of recognition targets are selected, or when the number of recognition targets is the same, such as the top N, when compared with other evaluation images, it can be treated as equivalent to the average confidence. it can.

  The evaluation image classifying unit 13 is a part that classifies evaluation images based on the average certainty factor calculated by the average certainty factor calculating unit 12. The evaluation image classification unit 13 holds a correspondence table between levels and average certainty as shown in FIG. 4, and the evaluation image classification unit 13 has the average certainty calculated by the average certainty calculation unit 12. It is classified by determining which level in the correspondence table is classified. The correspondence table in FIG. 4 is a table in which levels are divided into 8 levels.

  The evaluation index output unit 14 is a part that outputs a classification result (level) and an evaluation image in association with each other. For example, the evaluation index output unit 14 is a display or the like, and displays the classification result and the evaluation image in association with each other.

  As described above, the evaluation image processing apparatus 100 according to the present embodiment can classify the level of the evaluation image based on the certainty factor obtained by character recognition processing of the evaluation image.

  Next, the operation of the evaluation image processing apparatus 100 will be described. FIG. 5 is a flowchart of the evaluation image processing method showing the processing of the evaluation image processing apparatus 100. First, an evaluation image including characters is input by the evaluation image input unit 10 (S21). Then, the character recognition processing unit 11 performs character recognition processing on the input evaluation image according to the character recognition method designated by the operator among the plurality of character recognition methods set (S22).

  When the character recognition processing is performed by the character recognition processing unit 11, the certainty factor of the character recognition result is calculated by the character recognition processing unit 11 for each character obtained from the recognition result (S23). And the average certainty calculation part 12 calculates the average value per character of the certainty factor for all the characters obtained in the evaluation image, and the obtained average value is set as the average certainty factor of the evaluation image (S24). ). When the average certainty factor is calculated, the evaluation image classifying unit 13 classifies the evaluation image based on the average certainty factor (S25).

By such processing, the evaluation image processing apparatus 100 can perform evaluation image classification processing. FIG. 6 shows a specific example of an evaluation image evaluated and classified by the evaluation image processing apparatus 100 according to the present embodiment and its certainty factor. As shown in FIG. 6 (a), this evaluation image includes a signboard that reads “drinking water” in a certain landscape image. When the evaluation image processing apparatus 100 according to the present embodiment performs such evaluation image classification processing, a certainty factor is obtained for each extracted character as shown in FIG. In this evaluation image, characters that cannot be recognized in the human eye such as “,” and “(” and “●” are also extracted as extracted characters. The evaluation image processing apparatus 100 can classify the evaluation images by calculating the average certainty of the extracted characters as described above, as shown in FIG. When the average certainty factor based on the certainty factor is applied to the correspondence table of FIG. 4, the average certainty factor is (1 + 2 + 1 +... 3 + 7) /16=1.94.
Thus, this evaluation image can be classified as level 2.

  The character recognition method can be evaluated based on the classification result. For example, it is assumed that a certain number of evaluation images are extracted for each classified level, and the extracted evaluation images are evaluated for character recognition accuracy for each classification.

  Next, operational effects of the evaluation image processing apparatus 100 of the present embodiment will be described. In the evaluation image processing apparatus 100 of this embodiment, the character recognition processing unit 11 performs character recognition on the evaluation image input from the evaluation image input unit 10 according to the character recognition method designated by the operator, and recognizes each character. Calculate confidence for the result. Then, the average certainty calculation unit 12 calculates an evaluation value of the evaluation image based on the calculated certainty factor. In the present embodiment, the average value of the certainty factor is calculated as the evaluation value of the evaluation image, but the present invention is not limited to this. For example, a total value of the certainty factor may be calculated. Then, the evaluation index output unit 13 classifies the evaluation images based on the calculated evaluation value. Then, the evaluation index output unit 14 displays the classified evaluation image and its classification level. Thereby, it becomes possible to evaluate using a unique numerical index.

  In addition, it can evaluate from various angles by classifying an evaluation image into a plurality of levels based on an evaluation value. For example, a certain character recognition method is suitable for recognizing an image including a large number of characters, and another character recognition method can perform a recognition process with respect to an image including a small number of characters. For example, it is possible to grasp the performance of the character recognition method and the recognition characteristics. Therefore, it is possible to easily grasp the function-enhanced portion necessary for performance improvement.

DESCRIPTION OF SYMBOLS 10 ... Evaluation image input part, 11 ... Character recognition process part, 12 ... Average reliability calculation part, 13 ... Evaluation image classification part, 14 ... Evaluation index output part, 100 ... Evaluation image processing apparatus.

Claims (5)

In an evaluation image processing apparatus that evaluates a character recognition method,
An input means for inputting an evaluation image;
Character recognition processing means for performing character recognition on the evaluation image input by the input means, and calculating a certainty factor for a recognition result for each character;
Evaluation value calculation means for calculating an evaluation value of the evaluation image based on the certainty factor calculated by the character recognition processing means;
An evaluation image processing apparatus comprising: The evaluation value calculation means includes
2. The evaluation image processing apparatus according to claim 1, wherein an average value of the certainty factor per character is calculated as an evaluation value from the certainty factor of all characters in the evaluation image calculated by the character recognition processing unit. . The evaluation value calculation means includes
2. The evaluation image processing apparatus according to claim 1, wherein a total value of certainty factors of all or a predetermined number of characters in the evaluation image calculated by the character recognition processing unit is calculated as an evaluation value.   An evaluation image processing apparatus further comprising an evaluation image classification unit that classifies the evaluation image based on the evaluation value calculated by the evaluation value calculation unit. In the evaluation image processing method of the evaluation image processing apparatus for evaluating the character recognition method,
An input step for inputting an evaluation image;
Character recognition is performed on the evaluation image input by the input step, and a character recognition processing step of calculating a certainty factor for the recognition result for each character;
An evaluation value calculating step for calculating an evaluation value of the evaluation image based on the certainty factor calculated by the character recognition processing step;
An evaluation image processing method comprising: