JP3318588B2 - Local binarization method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an image processing method for obtaining a binary image from a multi-valued image.

[0002]

2. Description of the Related Art A conventional binarizing method of a multi-valued image will be described with reference to a flowchart of FIG.
In the prior art, a uniform threshold is obtained from a luminance histogram (step S102) for a multi-valued image (step S101) (step S103), and binarization is performed using the uniform threshold (step S104). An image has been obtained (step S105).

[0003]

However, when binarization is performed using a uniform threshold value for the entire screen, it is difficult to distinguish between "noise" and "thin figure".

[0004]

An image to be processed is divided to obtain a “cell” of n × n pixels, and m × n pixels are centered on the cell.
A measurement area of m pixels is set (n <m). Further, a local threshold value for the cell is obtained from a luminance histogram in the measurement area, and the local threshold value is used to divide the cell into two.
By repeatedly performing the process of converting the value into the cells of the entire region of the image to be processed, all the images to be processed are binarized. It should be noted that a limit value is set in advance for the local threshold value, and the extraction of noise components is suppressed.

[0005]

Embodiments of the present invention will be described below with reference to the drawings. The discriminant analysis binarization method is effective for data having a high bimodal luminance distribution, but there is a limit in image data or the like consisting of three elements. FIG. 1 shows a luminance histogram of image data in which a luminance component is composed of three elements. For example, if the threshold value is set to “dark”, “symbols” can be easily extracted, but if the threshold value is set to “light”, the image quality deteriorates rapidly, such as the “symbols” being crushed.
However, when the map data is directly viewed by a human, much information including the background can be extracted and recognized. This is because, as a feature of human vision, the sensitivity is reduced around areas with strong stimuli, and the sensitivity is increased by emphasizing the difference (differential) with surrounding areas for unknown parts. is there. Therefore, in the local binarization method according to the present invention, a binary image is acquired with the help of a human visual mechanism.

First, a luminance histogram of the entire target image is created, and a uniform threshold value is obtained using a discriminant binarization method. Next, the image to be processed is divided into “cells” of n × n pixels.
Is obtained, and a measurement area of m × m pixels which is slightly larger than the cell is set. Further, a luminance histogram of the measurement area is created, a local threshold value is obtained for the cell using a discriminant analysis binarization method, and the cell is binarized using the local threshold value. In order to suppress noise extraction, a threshold value is evaluated based on a preset limit value. That is, a uniform threshold value and a local threshold value are obtained for the entire image to be processed, and if the local threshold value is within the limit value, the cell is binarized using the local threshold value, If the value is outside the limit value, the cell is binarized using a uniform threshold value. This processing is repeatedly performed on the cells in the entire area of the image to be processed, and all the images to be processed are binarized to obtain a binary image.

FIG. 2 shows the local binarization method according to the present invention.
This will be described with reference to the flowchart of FIG. In the local binarization method according to the present invention, the processing target image (step S
A uniform threshold value is obtained using the discriminant binarization method (step S2), and the image to be processed is divided to obtain a “cell” of n × n pixels (step S3). A measurement area (m × m) is set at the center (step S4). Next, a luminance histogram of the measurement area is created (step S5), and a local threshold value for the cell is obtained (step S6). Next, the local threshold value is evaluated (step S7). If the local threshold value is within the limit value (YES), the local threshold value is adopted (step S8), and the cell (N × n) is binarized (step S10). Conversely, if the local threshold value exceeds the limit value (NO), a uniform threshold value is adopted (step S9), and the cell (n × n) is binarized (step S1).
0). This processing is repeatedly performed on the cells in the entire area of the image to be processed, so that all the images to be processed are binarized (step S11), and a binary image is obtained (step S12).

[0008]

As described above, in the local binarization method according to the present invention, an image to be processed is divided to obtain a "cell" of n × n pixels, and m × m pixels of the cell are centered on the cell. Set the measurement area (n <m). Further, a local threshold value for the cell is obtained from the luminance histogram of the measurement region, the cell is binarized using the local threshold value, and this process is repeatedly performed on the cells in the entire region of the image to be processed. By doing so, all the processing target images are binarized. Note that a limit value is set for the local threshold value in order to suppress noise extraction, and if the local threshold value is within the limit value, the cell is binarized using the local threshold value. If the value is outside the limit value, the cell is binarized using a uniform threshold value. In other words, by using the human visual mechanism, the sensitivity is reduced around the area with strong stimulation, and the unknown area is emphasized with the difference (differential) from the surroundings to increase the sensitivity and increase the multi-valued image. 2
In order to convert to a value, "dark figures", "light figures" or "noise"
A lot of information can be extracted and recognized from the image including the information. Therefore, the present invention relates to "dark figures" such as map data.
This is effective for obtaining a binary image from an image having a lot of information such as “thin figure” and “noise”.

[Brief description of the drawings]

FIG. 1 is a luminance histogram of an image in which a luminance component is composed of three elements.

FIG. 2 is a flowchart of a local binarization method according to the present invention;

FIG. 3 is a flowchart of a conventional binarization method;

──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Masahiro Ishikawa 1-9-9 Moto-Hongo-cho, Hachioji-shi, Tokyo Inside Chuo Denshi Co., Ltd. (72) Inventor Mitsuhiro Aoki 1-9-1, Moto-Hongo-cho, Hachioji-shi, Tokyo No. 9 Inside Chuo Denshi Co., Ltd. (72) Inventor Koki Ito 1-9-9 Motohongo-cho, Hachioji-shi, Tokyo Chuo Denshi Co., Ltd. (56) References JP-A-1-108684 (JP, A) JP-A-61-62979 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) H04N 1/40-1/409

Claims (1)

(57) [Claims] 1. An image processing method for binarizing an image to be processed, wherein a uniform threshold value is obtained for the entire region of the image to be processed, and the image to be processed is divided into “cells” of (n × n) pixels. , 1
A measurement area of (m × m) pixels is set around the cell (n <m), and a localization is performed on the luminance distribution in the measurement area using the discriminant analysis binarization method to binarize the cell. A threshold value is determined, a threshold value is evaluated based on a preset limit value, and if the local threshold value is within the limit value, the cell is binarized using the local threshold value, and the local If the threshold is outside the limit, the cell is binarized using the uniform threshold,
A local binarization method, wherein the processing target image is binarized by repeating this processing for all the processing target images.