JP3623998B2 - Image processing method and image processing apparatus - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to an image processing method and an image processing apparatus that can determine a keyword suitable for an input image.
[0002]
[Prior art]
Conventionally, in an image database equipped with a character recognition device, a character string obtained as a result of character recognition is for reuse when creating a document. And the file name was selected and entered by the user. In addition, there is a technique in which an image at a predetermined position of an input image is recognized and registered as a keyword.
[0003]
[Problems to be solved by the invention]
Therefore, when registering image data in the database, user processing such as keyword input is required, which makes it difficult to automate database registration. A method without adding a keyword is also conceivable. In this case, the user does not need to enter a key, but there is a problem that a simple search using a keyword becomes impossible because a keyword is not registered when searching later.
[0004]
[Means for Solving the Problems]
In order to solve the above problems, an image processing method of the present invention includes a character recognition step of generating a character string by character recognition processing of a character image in an input image, and a character string generated in the character recognition step. By extracting a plurality of words from the character string, an appearance frequency examining step for examining the appearance frequency of each word extracted in the extraction step, and an appearance examined in the appearance frequency examining step Based on the frequency, a keyword selection step for selecting a word to be used as a keyword of the input image from a plurality of words extracted in the extraction step, and a word selected in the keyword selection step together with the input image, A registration step registered in the database, and an image processing for controlling the image processing apparatus to execute each step. Further, in the keyword selection step, the maximum number of keywords that can be selected is set in advance, and when the keywords are selected in order from the word having the highest appearance frequency, the same number is selected. When there are a plurality of words having the appearance frequency and all the words having the appearance frequency are selected as keywords, the maximum number is exceeded, and the size of the character image of the word on the input image among the words having the appearance frequency In order from the largest, the maximum number is selected so as not to exceed.
[0005]
In order to solve the above problems, an image processing apparatus according to the present invention performs character recognition processing on a character image in an input image to generate a character string, and a character string generated by the character recognition unit. The extraction means for extracting a plurality of words from the character string, the appearance frequency investigation means for examining the appearance frequency of each word extracted by the extraction means, and the appearance examined by the appearance frequency investigation means Based on the frequency, a keyword selection unit that selects a word to be used as a keyword of the input image from among a plurality of words extracted by the extraction unit, and a database that selects the word selected by the keyword selection unit together with the input image The keyword selecting means is preset with a maximum number of selectable keywords. In addition, when selecting as a keyword in order from the word with the highest appearance frequency, if there are a plurality of words with the same appearance frequency and all the words with the appearance frequency are selected as keywords, the maximum number is exceeded. In such a case, it is characterized in that the words having the appearance frequency are selected so as not to exceed the maximum number in descending order of the size of the character image of the word on the input image.
[0014]
【Example】
The present invention will be described in detail with reference to the drawings.
[0015]
FIG. 7 is a block diagram showing the configuration of the apparatus in this embodiment. A CPU 1001 executes processing of the entire apparatus, and controls determination and processing in accordance with a control program stored in the ROM 1002 or the RAM 1003. Reference numeral 1002 denotes a ROM which stores a control program of a flowchart described in the present embodiment or data such as predetermined parameters used for processing. A dictionary used for character recognition processing and word separation processing is also stored in the ROM 1002. Reference numeral 1003 denotes a RAM, which includes a working memory area for storing data being processed by the CPU 1001. When a control program for processing executed in this embodiment is supplied from a removable storage medium such as an FD, the control program is stored in the RAM 1003. The database created in this embodiment is also stored in the RAM 1003. Reference numeral 1004 denotes a scanner which optically reads a document image. The read image data can be stored in the RAM 1003. Reference numeral 1005 denotes a keyboard which can input various codes and operator instructions. Reference numeral 1006 denotes a pointing device, which can indicate a desired position on the display screen of the display 1007, and can input a selection or cancellation instruction by clicking a button. Reference numeral 1007 denotes a display, which includes a CRT or a liquid crystal display. 1008 is a printer such as an LBP or an ink jet printer, 1009 is an external storage device such as a floppy disk, and a floppy disk drive (FDD) for reading and writing data 1010 is a data transfer between these components This is a data bus for performing.
[0016]
FIG. 5 is a diagram showing a functional configuration of the image database system according to the embodiment of the present invention. In this system, the input image of the document input from the scanner 1004 is first subjected to image region separation processing in the image region separation means (501), and then a character string as a recognition result is generated by the character recognition means (502). The separating means (503) analyzes the character string of the recognition result while referring to the dictionary (506) and separates it for each word. The keyword determining means (504) performs keyword determination, and the keyword selected as a result is input together with the input image. Register in the database (505).
[0017]
FIG. 1 is a flowchart showing overall processing from image input to registration in a database of an image database system according to an embodiment of the present invention. First, the image region separation unit 501 analyzes an input image, separates a portion with character or image data into a rectangle so as to be surrounded by a rectangle, and classifies each block according to the image characteristics of the block. However, here, each block extracted from the input image is classified into a character area or an image area. That is, for example, if the input image is the image shown in FIG. 2, several blocks B301 to B307 in FIG. 3 are extracted as a result of the processing in S101, and each block is classified, and B301 to B306 are character regions, B307 is determined to be an image area. As a result, a plurality of image blocks existing in the input image are taken out and separated into a character part image area and other parts, that is, an image part image area, character part image area information is stored in the RAM 1003, and is passed to the character recognition means 502 ( S101).
[0018]
Next, character recognition is performed on each image area determined as a character part image area by the character recognition unit 502. First, the image of the character part image area is further divided into character units (character cutting process). In the character cutting process, the image of each character image region is first cut out in the row direction (left and right direction) to cut out the row position, and then the histogram is cut up and down in each row to cut out the position for each character. For example, when character cut processing is performed on the character part image area B302 in FIG. 3, the character part image areas B401 to B409 in FIG. 4 are separated. A pattern matching process is performed on each one-character image area using a dictionary stored in the ROM 1002, the character represented by each one-character image area is recognized, and a character string corresponding to the character is output (S102). To store. For example, as a result of character recognition for each block in FIG.
・ "Important documents" (from B301)
・ "Kannon Co., Ltd. New Product Briefing" (from B302)
・ "Kannon Co., Ltd. President and Representative Director Satoshi Daibutsu" (from B303)
・ "Dear Sirs, I would like to thank you for your success. Now, I will inform you that we have planned a new product briefing for the launch of our new product." (From B304)
・ "Note" (from B305)
・ "Date and time: XX month x day 8:00 am to 5:00 pm Location: △△ prefecture ◎◎ City 1-2-3 Special venue (see map)" (from B306)
Will be generated.
[0019]
Next, the character string resulting from S102 is read from the RAM 1003 and divided into words by the word separation means (503) (S103). The word separation means separates the recognition result character string for each word (morphological analysis) while referring to the dictionary (506) (S103) and stores it in the RAM 1003. When storing in the RAM 1003, information on the part of speech of each word is extracted from the dictionary 504, added to the word, and stored.
[0020]
Next, the keyword determination unit 504 performs keyword determination processing on all the words extracted from the input image by the word separation unit (503) stored in the RAM 1003 (S104).
[0021]
The detailed processing flow of the keyword determination processing (S104) is shown in the flowchart of FIG. 6 and will be described with reference to this flowchart. First, the part-of-speech information added to each word stored in the RAM 1003 is determined, and narrowing down by part-of-speech is performed. In the present embodiment, nouns are extracted as being suitable for keywords (S601). For example, the following words are extracted as nouns from the character string that is the recognition result of each of the previous blocks.
[0022]
・ "Important documents" (from B301)
・ “Kannon Co., Ltd.”, “New Product Information Session” (B302)
・ "Kannon Co., Ltd.", "President and CEO", "Great Buddha" (B303)
・ "Seiei", "Okay", "Our new product", "Release", "New product briefing", "Planning", "Guide" (B304)
・ "Date and time", "Place", "△△ prefecture ◎◎ city", "special venue", "map reference" (B305)
The word is raised. Next, it is assumed that the word having the highest appearance frequency is more important in the document, so the appearance frequency of each of these words is examined (S602). Then
・ Kannon Co., Ltd. (Frequency 2, from B302 and B303)
・ New product briefing (Frequency 2, from B302 and B304)
・ Important documents (Frequency 1, from B301)
・ President and CEO (Frequency 1, B303)
・ Great Buddha (frequency 1, from B303)
・ Seiei (Frequency 1, from B304)
・ Good night (frequency 1, from B304)
・ Our new product (frequency 1, from B304)
・ Release (frequency 1, from B304)
・ Planning (frequency 1, from B304)
・ Guidance (Frequency 1, from B304)
・ Date (frequency 1, from B305)
・ Location (Frequency 1, from B305)
・ △△ prefecture ◎◎ city (frequency 1, from B305)
・ Special venue (frequency 1, from B305)
・ Map reference (frequency 1, from B305)
The result is obtained.
[0023]
In the next step S603, the maximum number (K) of keywords that can be set in one input image is assumed to be five here. Further, the maximum value (N) of the appearance frequency of nouns is 2 from the result of the previous step S602. Therefore, as a result of processing in this step,
K = 5, N = 2, S = 0, I = 2
It becomes. Here, S represents the total number of selected nouns, and I is a work variable.
[0024]
In the next step S604, the number of nouns whose appearance frequency is I is checked and set to M. Therefore, here, since there are two nouns with the appearance frequency 2, “Kannon Co., Ltd.” and “New Product Information Session”, the result M of this processing is 2.
[0025]
Next, it is determined whether or not the sum of S and M is equal to or less than K, that is, whether or not a keyword having an appearance frequency of I is equal to or less than the maximum number of keywords (S605). In this example, since S = 0 and M = 2 here, the determination result in step S605 is true when S + M = 2, and the process continues to step S606, where the noun with the appearance frequency I (= 2) is internally processed. To the selected state. The word to be selected is stored in the keyword selection word storage unit in the RAM 1003. Here, since M nouns are selected, S is increased by M (S607). In the case of this example, the value of the result S in step S607 is 2 here.
[0026]
Next, it is determined whether I is 1 (S608). If I is 1, it is determined in step S604 that the noun has an appearance frequency of 1, that is, there is no more noun to be examined, and the process moves to step S613. In this example, since I is 2 here, the result of the determination is false, the process proceeds to step S609, I is decremented by 1 and becomes 1, the process returns to step S604, and the appearance frequency is 1. A process of selecting a word to be registered as a keyword from words is started.
[0027]
This time, since the word having the appearance frequency of 1 is a target, in this example, in step S604, M is 14 here, and the determination in the next step S605 is false (S + M = 16, K = 5), The processing moves to step S610.
[0028]
In step S610, it is checked whether the number of nouns currently selected is the same as the limit on the number of keywords. If they are the same, the next step S611 and step S612 are skipped, and the process proceeds to step S613. In this example, since S = 2 and K = 5, the determination result is false, and the process proceeds to step S611.
[0029]
In step S611, (KS) keywords are selected from nouns having the appearance frequency I. This is because S nouns are selected as keywords based on frequency information, but since K keywords can actually be registered, the remaining (K-S) keywords can be registered, so the remaining unselected keywords This is to select from nouns. In this example, since K = 5 and S = 2, 5-2 = 3 keywords are selected from the remaining 14 nouns.
[0030]
The remaining (KS) keywords are selected by selecting (KS) items in the order in which the nouns appear in the direction of the sentence (in this case, in this example, “important documents”, “representative director president” In the present embodiment, the character recognition unit 502 did not output information on the character size, but the character recognition unit 502 does not output the size of each character. And the remaining (KS) nouns may be selected in the descending order of the characters, or these may be combined.
[0031]
Next, since (K-S) keywords are selected in step S613, the selection process of all K keywords is completed, and the selected S nouns are set as keywords.
[0032]
Finally, the keyword selected in the keyword determination process (S104) and stored in the keyword selection word storage unit of the RAM 1003 is registered in the database in the RAM 1003 together with the input image (S105), and the process ends.
[0033]
A word registered as a keyword is displayed in a table of contents when searching for an image from the database, or when a word is specified for search, an image stored in the database together with the same word as that word is displayed or displayed. Since it is printed, image retrieval is facilitated.
[0034]
In this embodiment, some keywords are selected by the keyword selection means due to the limitation of the number of keywords. However, the keywords selected with the selection number set to one here are stored in the database as file names of input files. You may register.
[0035]
In this embodiment, all the characters are recognized and then the words are extracted. For example, in the case where words such as English are separated by spaces, matching is performed between the images in the word area before the characters are recognized. To detect a word having a high appearance frequency, and recognize the detected word as a character and register it as a keyword.
[0036]
【The invention's effect】
As described above, according to the present invention, since a word having a high appearance frequency in the input image is selected as a keyword to be added to the input image, it is possible to select an accurate keyword without the manual operation of the operator using the keyboard. Can do.
[0037]
As described above, according to the present invention, since character recognition is performed after selecting a word pattern to be registered as a keyword, processing time until selecting a keyword is eliminated without spending extra time for character recognition processing. Can be shortened.
[0038]
As described above, according to the present invention, since words of a specific part of speech are narrowed down as keywords for the keyword selection process, the selection process time can be shortened and a word suitable as a keyword can be selected. .
[0039]
As described above, according to the present invention, since the selection of words to be stored together with the image input by the scanner is extracted from the input image, an operation using the keyboard is not necessary, and the operability is improved.
[Brief description of the drawings]
FIG. 1 is a flowchart showing the overall flow of processing of the present embodiment.
FIG. 2 is a diagram illustrating an example of an input image.
FIG. 3 is a diagram illustrating an example of an image region separation process result.
FIG. 4 is a diagram illustrating an example of a single character image separation process result.
FIG. 5 is a functional configuration diagram of the apparatus according to the present embodiment.
FIG. 6 is a flowchart showing the flow of keyword determination processing.
FIG. 7 is a configuration block diagram of the apparatus according to the present embodiment.

Claims (7)

A character recognition step for generating a character string by performing character recognition processing on the character image in the input image;
An extraction step of extracting a plurality of words from the character string by performing morphological analysis on the character string generated in the character recognition step;
An appearance frequency examining step for examining the appearance frequency of each word extracted in the extraction step ;
A keyword selection step of selecting a word to be used as a keyword of the input image from a plurality of words extracted in the extraction step based on the appearance frequency examined in the appearance frequency investigation step;
A registration step of registering the word selected in the keyword selection step together with the input image in a database, and an image processing method for controlling the image processing apparatus to execute each step,
In the keyword selection step, when the maximum number of keywords that can be selected is set in advance and words are selected in order from the word with the highest appearance frequency, there are a plurality of words having the same appearance frequency. And, when all the words with the appearance frequency are selected as keywords, the maximum number is exceeded, the words in the appearance image in order of the size of the character image of the word on the input image from the largest An image processing method for controlling an image processing apparatus , wherein the selection is made so as not to exceed a maximum number . Words extracted by the extraction step, the image processing method according to claim 1, which is a word in a particular part of speech. The image processing method according to claim 2, wherein the specific part of speech is a noun. The image processing method according to claim 1, further comprising a display step of displaying the words registered in the database together with the input image in a table of contents. 2. The image processing according to claim 1, further comprising: a search step of reading out and outputting an input image registered together with the word when the word specified by the search is registered in the database. Method. The image processing method according to claim 1, wherein the input image is image information input from a scanner. Character recognition means for generating a character string by character recognition processing of a character image in the input image;
Extraction means for extracting a plurality of words from the character string by performing morphological analysis on the character string generated by the character recognition means;
Appearance frequency investigation means for examining the appearance frequency of each word extracted by the extraction means ;
Keyword selection means for selecting a word to be used as a keyword of the input image from a plurality of words extracted by the extraction means based on the appearance frequency checked by the appearance frequency search means;
A registration means for registering a word selected by the keyword selection means together with the input image in a database, and further comprising:
The keyword selection means is preset with a maximum number of keywords that can be selected, and there are a plurality of words having the same appearance frequency when selecting the keywords in descending order of the appearance frequency. And, when all the words with the appearance frequency are selected as keywords, the maximum number is exceeded, the words in the appearance image in order of the size of the character image of the word on the input image from the largest An image processing apparatus that is selected so as not to exceed the maximum number .

Images