JPS60126782A - Character recognizing device - Google Patents

Abstract

PURPOSE:To decrease the storage capacity of a dictionary and also to improve the recognizing rate and recognizing speed by extracting stably the feature of a stroke at a stroke detecting section to decrease kinds of features derivated from one character. CONSTITUTION:A picture data read from an original 10 is binary-coded by a photoelectric converting section 20 and a pre-processing section 30 applies processing such as character segmenting and normalization from a coded picture data. Moreover, a feature extracting section 40 detects a feature amount from a character pattern, a retrieval section 50 compares the extracted feature amount with the feature amount of a known character stored in a dictionary storage section 60 in advance, and the result of comparison is outputted to an output section 70. A character pattern from the pre-processing section 30 is added to a stroke feature extracting block 42 and an entire feature extracting section 41 of the extracting section 40. Then a direction coding section 43 of the block 42 codes the character pattern by directions, the result is processed by the post expanding section 44 and stroke feature extracting section 47 and the feature of the stroke is extracted stably by using the feature from the extracting section 41.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field This invention relates to character recognition devices, particularly those that recognize characters based on the characteristics of their strokes. Binarize the multivalued image, extract stroke features for each direction based on this binarized code information, calculate the feature amount for each direction from this extracted stroke information, and calculate the feature amount for each direction based on this % trace amount. The system is configured to perform recognition based on the following information. In this case, among the extracted strokes, connected strokes are extracted separately.

By the way, in conventional character recognition devices of this type, for example, all code patterns are compared before and after stroke extraction in (a) and Φ) in Figure 1 and (a) and (b) in Figure 2, respectively. As shown in Figure 2, even if the patterns were very similar to each other, the number of strokes was often extracted differently. In FIGS. 1 and 2, (a) shows the code pattern of the straw pm delivery and the code pattern extracted from the Φ fly stroke, respectively. For this reason, the above feature values are not stable and various features are derived for the same character, which increases the storage capacity of the dictionary required for recognition and causes a decrease in recognition rate and recognition speed. This caused the dog to get angry.

Purpose This invention solves the above-mentioned problem. It is possible to stably extract stroke features, thereby reducing the number of features derived from the same character and improving the dictionary. An object of the present invention is to provide a character recognition device capable of reducing storage capacity and improving recognition rate, recognition speed, etc.

Structure The structure of the character recognition device according to the present invention is such that the character recognition device performs recognition by binarizing an optically read character pattern and then extracting the percent signature of the stroke. A means for encoding each direction, a stroke extraction unit that extracts stroke information for each direction from the information coded for each direction, and determining whether or not to integrate strokes based on the extracted stroke information and integrating the strokes. The stroke selection section and the stroke extraction section i are provided with a stroke feature extraction section that extracts features based on information from the stroke selection section, and operate the stroke selection section based on the information from the stroke extraction section. Thus, the information adopted by the stroke feature extraction section can be controlled whether to obtain it directly from the stroke extraction section or through the entire stroke selection section. Features.

Embodiments of the present invention will be described below with reference to all the drawings.

FIG. 3 shows the configuration of a character recognition device according to an embodiment of the present invention divided into functional blocks.
Preprocessing section (9), feature extraction section 4o, search section father, dictionary storage section 60. and an output section 7o. In FIG. 3, an input document 10 is converted into binary code image data by photoelectric conversion $20. The preprocessing unit 30 extracts characters from the photoelectrically converted image data and performs smoothing (
The 11% feature extraction unit 40, which performs processing such as smoothing processing and normalization, extracts feature amounts from the character pattern and sends them to the search unit. The search unit 50 calculates the percentage of the extracted feature amount and the superior known characters stored in the dictionary storage unit 60 in advance.
The trace amount and the total amount are compared with each other, and the literary journal 4 is performed based on this, and the result is sent to the output section 70. The output unit 70 outputs the result.

FIG. 4 shows the configuration of the percentage feature extraction section 60? show.

The feature extraction section 40 shown in the figure is composed of an overall feature extraction section 41 and a stroke feature extraction block 42. Furthermore, the stroke % image extraction block 42 includes a direction-based coding section 43, an extension section, a stroke extraction section 45, y:,
Tortoise selection section 46 and stroke feature extraction section 4
7.

In FIG. 4, the character pattern sent from the preprocessing unit 3o is input to the overall feature extraction unit 41 and the stroke feature extraction block 32.The stroke feature extraction block 42 processes each stroke of the input character pattern. In addition, the overall feature extraction unit 41 extracts features such as the number of holes in the input character pattern and the center of gravity of the two central characters.Here, the operation of the stroke feature extraction block 42 will be explained as follows. First, in the directional code 16 section 43, an extension process is performed to extend all strokes of the arrow a, where the input character pattern is coded in each direction, using an extension member. This is done by extending the stroke as long as there are black pixels in the direction indicated by the separately coded code.

The character pattern that has been decompressed as described above is
Is the stroke information sent to the stroke extraction section 45 and the stroke information from the stroke extraction N545 sent to the outside of the stroke feature extraction section 47 via the stroke selection section 46?
A certain - is sent directly to the stroke feature extraction section 47 without passing through the stroke selection section 46'. Here, as shown in FIG. 5, the stroke extraction section 45, the stroke selection section 46, and the stroke feature extraction section 47 are connected to each other by making a connection table.The connection table 48 detects the presence or absence of connected strokes. The configuration is such that the presence or absence of the connected stroke is indicated by a flag.

Now, the stroke extraction section 45 extracts the strokes of the character pattern that has been expanded in each direction. This stroke m is determined by checking all the connection states of black pixels in each direction, and for patterns such as those shown in (a) and Φ) in Fig. 6, by performing stroke separation with reference to the background information. Collect data for stroke feature extraction such as center coordinates, length and width.

Then, if there is a connected stroke, the position information and connection state are written in the connection table, and the connection flag t is activated at the same time.The low stroke selection unit 46 checks the state of the connection flag and confirms that the flag is not active. In the active state, the stroke 悄@ from the stroke extraction section 45 is sent as is to the stroke feature extraction section 47. Further, as soon as the connection flag is in the active state, the stroke selection section 46 starts operating,
All the following processes are performed on the connected strokes.

(1) Regarding the direction code as shown in FIG. 7, direction 1 is on the X axis shown in FIG. 6, direction 2 is on the X axis shown in FIG. 3. 4 are projected onto the Y axis shown in the figure.

(2) Let the respective lengths on the projection axis be 't-LI and L2, and let the length of the electric part be L3, then (L1+
L2-2・L3 )/(L1+L2-L3)>TH
(predetermined threshold value) In the case of a connected stroke, stroke extraction 845
The stroke information from is directly transmitted to the stroke feature extraction unit 48. (3) (L1+L2-2・L3) / (Id+L2-
I, 3) For strokes where ≦TH, direction 1
If so, the projection is taken in the Y-axis direction, and the direction 2. 3.
If it is 4, the projection is taken in the X-axis direction. □ (4) Similar to (2), the degree of overlapping of the projections is fully determined.

(L1+L2-2- L3) / (L1+L2-L3
), the stroke information t from the stroke extraction section 45 is adopted as input information to the stroke feature extraction section 47.

Also, (L1+L2-2・I, 3) / (L1+L2
-1,3)≦TH, the strokes are integrated.

The above processing operation sound flowchart is as shown in FIG.

Further, FIG. 9 shows how to create a concatenation table for the pattern of FIG. 6.

No. 10- is the stroke selection section 4 in another embodiment.
6 shows an operation flowchart of No. 6. It would take time to take the projection as described above, but here, as the criterion for (3) above, we will use the center coordinates (X
i, yl), (X2. y2) - In other words,
For direction 1, 1yl-y21, and for direction 2. 3.
4, 1xl-x21 is the determined TI
(Judgment is made based on whether it is larger or smaller.) ゛Effects As described above, the character recognition device according to the present invention can prevent the unstable appearance of strokes in its stroke extraction section, thereby creating stable features. Since the amount can be obtained, it is possible to reduce the number of types of features derived from the same character and reduce the dictionary capacity.
It is possible to obtain effects such as reducing the total cost, increasing the recognition rate, and improving the recognition speed.

[Brief explanation of drawings]

FIGS. 1 and 2 each show an example of the operation of the conventional device, and are diagrams showing a comparison of stroke patterns coded by direction and extracted stroke patterns, and FIG. 3 shows characters according to the present invention. FIG. 4 is a block diagram showing an embodiment of the recognition device. FIG. 4 is a block diagram showing the configuration of the feature extraction section in FIG. 3. FIG. FIG. 6 is a diagram showing a complete comparison of stroke patterns coded by direction and extracted stroke patterns in the apparatus according to the present invention, FIG. 7 is a diagram showing direction codes, and FIG. Fourth
FIG. 9 is a diagram showing an example of the contents of the concatenation table in FIG. 5, and FIG. 10 is a flowchart showing another embodiment of the present invention. ...Original structure 20...Light' [conversion section 30...
・Pre-processing unit 40 - Feature extraction unit 50... Search unit
Mark... Dictionary storage unit 70... Output unit 41... Overall feature extraction unit 42... Stroke feature extraction block
43... Directional coding member... Extension part 45...
- Stroke extraction section 46... Stroke selection section 4
7... Stroke feature extraction unit 48... Connection table Applicant's agent Kiyoshi Torii Figure 6 Figure 9 Figure 1O

Claims (3)

[Claims] (1) Optically read character pattern t2i Lr:
After that, the characteristics of the stroke are extracted and recognized, and in a 5-character recognition device, a means is used to code the binarized character pattern by direction, and stroke information is extracted by direction from the information encoded by direction. stroke extraction section,
A stroke selection section that determines whether or not to integrate strokes based on the extracted stroke information and performs the integration of strokes, and -1 and the stroke selection section based on the information from the above-mentioned Stro=z Totabe or the above-mentioned stroke selection section. A stroke feature extraction section for extracting features is provided, and the operation of the stroke selection section is controlled based on information from the stroke extraction section, so that the information adopted by the stroke feature extraction section is used to select the strokes. A character recognition device characterized in that seven lines are obtained either directly from the extraction section or via the stroke selection section. (2) The stroke selection section takes projections of the connected strokes onto the horizontal and vertical axes, and the partial forces that do not overlap in the projections of the respective strokes; and those that do not exceed a predetermined threshold on both axes. 2. The character recognition device according to claim 1, wherein the character recognition device is configured such that the strokes are integrated, and all the information of the stroke extraction section that is not integrated is directly adopted by the stroke feature extraction section. (3) The stroke selection unit is configured as an IC to determine conditions for integrating strokes based on the difference in center coordinates of each stroke. Character recognition device described in section -