JPH03192476A - Binarizing method for picture - Google Patents

Abstract

PURPOSE:To prevent a pattern from being erroneously recognized by nonuniformity by applying approximately averaged intensity distribution for the unit of a small area to a threshold value itself to be the reference of binarizing. CONSTITUTION:A differentiation processing is executed upon an inputted picture A and a differentiated picture B is binarized at a certain density level so as to obtain an edge picture C of the pattern. By executing an enlargement processing upon the edge picture C, a binary picture D is obtained slightly larger than the pattern part. Next, the extraction of a maximum value near each position is executed upon the input picture A and a picture E is obtained. Then, the extraction of a minimum value is executed similarly and a picture F is obtained. A picture G is made with the density level between the density levels of the maximum value picture E and the minimum value picture F and compared with the original input picture A as a threshold picture and only concerning the area of the picture D which are is slightly larger than the pattern part, a binarizing processing is executed. Thus, even when there is the nonuniformity of light intensity or printing density, etc., the pattern part can be exactly separated from a background part and the picture can be binarized.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a technique for binarizing an input image, which is necessary when performing pattern measurement using image processing.

[Prior Art] Generally, when binarizing a two-dimensional image input through an image input device such as a television camera, a single threshold value is set for the entire image, and Binarization was performed depending on whether the level of the input image signal was high or low.

[Problems to be Solved by the Invention] When a television camera is assumed as an image input device, it is conceivable that the illumination of the object is not uniform and there is uneven brightness. Further, even if the illumination is uniform, when reading a printed matter, the print density may not be uniform throughout.

If you input such an image and perform binarization processing by setting a single threshold value for the entire image, the background part may become a pattern part, or vice versa. It is easy to guess what will happen.

[Means for solving the problem] Using image differentiation processing and enlargement processing, the input image is roughly separated into a background part and a pattern part, and this pixel is divided into a small area area of the input image, for example, pixel by pixel. In addition to extracting the maximum intensity value and minimum intensity value in a group of pixels including the vicinity of the center, a variable threshold value is set for the entire image at an intensity level located between these two values, and based on this variable threshold value, The input image signal is binarized.

[Effects] Even if there are irregularities in illumination, print density, etc. in the image input object, the threshold itself, which is the standard for binarization, is given an intensity distribution roughly averaged in small area units. Compared to the conventional method in which a uniform threshold value is set for the entire image, misrecognition of patterns due to the above-mentioned unevenness is less likely to occur.

[Embodiments] Hereinafter, the present invention will be explained in detail based on embodiments shown in the drawings.

FIG. 1 is a flow diagram showing the processing procedure of the method of the present invention.

Here, a case will be described in which the density level of the pattern part is higher than that of the background part, but the present method can be easily applied to the opposite case as well. In FIG. 1, the names of images are represented by capital letters A...H.

FIG. 2 shows an example in which the characters rABCJ are input as an example of the image input object 10, and shows the density intensity on a line 21 that crosses this input object 10.

The English capital letters A...H in Figure 2 are the images A... in Figure 1.
It corresponds to H.

The image processing procedure for the above example will be described below.

First, an image A input through an image input device such as a television camera is subjected to differential processing to obtain a differential image B.

This differentiation process removes the background part in the input image,
Since an image B in which the edge portion of the pattern is emphasized is obtained, this image is binarized at a certain density level to obtain an edge image C of the pattern. In image C, the edge portion of the image is extracted without being affected by the background portion or uneven illumination of the image. By performing enlargement processing on this edge image C, it is possible to obtain a binary image slightly larger than the pattern portion.

In this way, input image data can be roughly separated into a background part and a pattern part.

In FIG. 2, the intensity distribution on the line 21 is
It is shown for D.

Next, a method of setting a variable threshold value will be explained. For input image A, one image E is obtained by extracting the maximum value near each position. The specific method for extracting this maximum value is explained in the third section.
The diagram will be explained.

We will take 3×3=9 pixels as the above “neighborhood”,
In FIG. 3, nine pixels a, b', c, . . . , i centered on a specific pixel e on the line 21 in FIG. Then, the maximum value in the vicinity of the specific position e, that is, a in the example of FIG. 3.

Extract the maximum value among the 9 pixels of b, c...i, associate this maximum value with the position of e, and save it to e゛. Repeat the same operation for all pixels to correspond to each pixel. A maximum value image 32 consisting of a matrix of maximum values a', b', C', .

The image E in FIG. 2 shows the neighboring maximum value image on the line 21, that is, the intensity distribution of dl, ef', . . . on the line in the example of FIG.

Similarly, the minimum value is extracted to obtain image F.

It is clear that the optimal binarization level is between the density levels of the maximum value image E and the minimum value image F.

Therefore, an image G having a density level between the density levels of the maximum value image E and the minimum value image F (for example, the center value) is created, and this image G is used as the image used to determine the threshold value for binarization. do. The image G has different density levels depending on the location, making it possible to determine the binarization level depending on the location.

Image G is used as a threshold image and compared with the original input image A, and binarization processing is performed based on the magnitude relationship. However, if this is done alone, there may be areas in the background area that have a higher density level than the threshold image G, so it is not possible to perform binarization processing only on areas of the image that are slightly larger than the pattern area. desirable.

[Effects of the Invention] According to the present invention, even if the image input object has unevenness in illumination intensity, print density, etc., the pattern portion to be recognized in the input image can be separated from the background portion without being affected by this. It is possible to accurately separate and binarize.

[Brief explanation of drawings]

FIG. 1 is a flow diagram showing the processing procedure according to the present invention. FIG. 2 is a diagram showing the intensity distribution on the line of the input image and each intensity distribution on the line of the image obtained by image processing,
FIG. 3 is a diagram illustrating maximum value extraction. 10... Input target, 21... Input line, A...
Input image, B...differential image, C...pattern edge image, D...enlarged pattern binary image, E...maximum value image, F...minimum value image, G...threshold value Image, H.
...Binarized image based on comparison with G. ■

Claims (1)

[Claims] When binarizing a signal input by an image reading device such as a television camera, the input image is divided into background parts and patterns by performing differential processing, enlargement processing, etc. on the input image. The maximum intensity value and the minimum intensity value within each small area of the input image are extracted, and a variable threshold value is set for the entire image at an intensity level located between the two values. An image binarization method, comprising: binarizing an input image signal based on the threshold value.