CN1941838A - File and picture binary coding method - Google Patents

Abstract

The present invention provides an image processing method for binarizing a document image, which includes the following steps: a) in the global thresholding processing, determining a global threshold for binarizing an image, and dividing the global threshold according to the global threshold The pixels of the document image are divided into three categories: black, white and undetermined pixels; b) determine an adaptive binarization threshold for each undetermined pixel, and binarize the undetermined pixels according to the adaptive binarization threshold .

Description

Document Image Binarization Method

technical field

The invention relates to the field of image processing, and specifically provides a technology for converting digital images obtained from scanners, fax machines or digital cameras into binary images. Fields of application of the invention are document image processing, document management and document recognition.

Background technique

In contemporary society, documents are the primary information carrier. Therefore, the present invention improves the binarization of images, especially document images composed of text, tables, lines and pictures. Since the information of a document image is essentially binary information, under ideal conditions, it can be represented by a single foreground and background, for example, white represents the background, and black represents useful information, that is, the foreground. However, in practical applications, due to the printing process, uneven reflection, the diversity of the content of the document itself, and various rich artistic effects, the foreground and background in the image usually change. The purpose of document image binarization is to separate useful information from useless information and represent the result as a binary image.

Image binarization is a necessary step in many applications. For example, US Patent No. 5,452,107 proposes a method to determine the binarization threshold based on the density of the local area of the original image, including the average value of the target pixel and surrounding pixels. The disadvantage of this method is that only limited information can be provided locally.

Contents of the invention

The object of the present invention is to provide a document image binarization method that can solve the above-mentioned problems in the prior art.

In order to achieve the above object, the present invention provides an image processing method for binarizing a document image, comprising the following steps: a) in the global thresholding processing, determining a global threshold for image binarization, according to the The global threshold divides the pixels of the document image into three categories: black, white and undetermined pixels; b) determine an adaptive binarization threshold for each undetermined pixel, according to the adaptive binarization threshold, the Pending pixel binarization.

The document image binarization method of the present invention combines global and local information, and effectively utilizes image local information and historical information at the same time, so it can provide higher quality binarized document images.

Description of drawings

The above and other objects and features of the present invention will become clearer through the following description in conjunction with the accompanying drawings, wherein:

Fig. 1 outlines the flowchart of the image binarization method proposed by the present invention.

Fig. 2 shows a detailed flowchart of the preprocessing module of the method of the present invention in Fig. 1 .

FIG. 3 shows a detailed flowchart of the global thresholding module of the method of the present invention in FIG. 1 .

Fig. 4 shows an example of a document image histogram after global thresholding, and correspondingly marks three global thresholds T1, T2 and T3 obtained by the global thresholding method.

FIG. 5 shows a detailed flowchart of the local thresholding module of the method of the present invention in FIG. 1 .

FIG. 6 shows a detailed flowchart of the post-processing module of the method of the present invention in FIG. 1 .

Fig. 7 shows an example of the process of binarizing an image by applying the method of the present invention.

Detailed ways

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

If the pixel values or color values of the background and useful information (or called the foreground) in the image are consistent in the whole image, then a high-quality binarized image can be obtained by using a single threshold. This approach is called global thresholding.

However, most of the document images currently used are rich in diagrams and artistic effects, and a single threshold often introduces noise or fails to retain useful information. The method of binarizing different pixels or pixels in different regions with different thresholds is usually called local thresholding.

Fig. 1 outlines the flowchart of the image binarization method proposed by the present invention. The image binarization method of the present invention is carried out in combination with global and local information.

Referring to Fig. 1, in the document image binarization method of the present invention, its input is a paper or electronic document 10, after the preprocessing module 11, the global thresholding module 12, the local thresholding module 13 and the postprocessing module 14 is transformed into an electronically binarized image.

If the input document 10 is a paper document, it needs to use an optical scanning device such as a scanner, a fax machine or a digital camera to convert it into a digital image that can be processed by a computer. The format of the digital image can be BMP, JPEG, TIF, etc.

The preprocessing module 11 performs a series of processing on the image which will be described below, and the processing result is used by the subsequent thresholding module.

Thereafter the global thresholding module 12 determines two thresholds to classify image pixels into white, black and undetermined pixels. Pending pixels refer to the set of pixels whose classification cannot be determined according to the histogram information in the global threshold stage. These pixels may be artistic effects, charts, photos, tables or even text information. Since global thresholding can process most of the image pixels, it can significantly improve the speed of binarization. Another benefit is that since the global thresholding stage does not distinguish between foreground and background, it can preserve the anti-color information in the document, that is, the color of the text symbol is darker than the background color.

In the present invention, the local thresholding module 13 determines a binarization threshold for each undetermined pixel according to the local features of the image and historical threshold information. The local features here include the mean and variance of the local area of the image. The historical threshold information comes from neighboring pixels that have been binarized. In the present invention, the use of historical threshold information is very important, which can significantly improve the quality of the output binarized document image.

Finally, the post-processing module 14 processes the image binarized by the global thresholding module 12 and the local thresholding 13 to remove noise on the image. Generally speaking, there are three types of such noise: sticking of text strokes, disconnection of text strokes, and isolated noise points. The post-processing method of the present invention can remove most of the noise in the image without introducing new noise.

After the above processing, the effective information of the input document 10 is represented as a binary document image 15 . This image can be used in many fields, such as further image analysis, color detection of text words, image compression of documents, layout analysis of documents, and optical character recognition, etc.

Each module in Figure 1 is described in detail below through Figure 2-6.

FIG. 2 shows the flow of the preprocessing module 11 in detail. The function of the preprocessing module 11 is to smooth the image to remove noise, and at the same time provide necessary data for the subsequent global thresholding module 12 . If the input is a paper document, it is first digitized by module 101 to generate a digital image. If it is a color image, it is converted into a grayscale image through the module 102 or each channel is processed separately. According to the content and quality of the image, the histogram equalization module can be used to process the gray scale. The subsequent low-pass filter 104 may be chosen to be a linear filter such as a Gaussian filter, or a non-linear filter such as a mean filter.

After that, the image is divided into image blocks. If the difference between the maximum and minimum values of pixels in the image block is less than the preset threshold, the image block is considered to be uniform, which cannot provide meaningful information for determining the global threshold. Therefore, in the image In the mask estimation module 105, the uniform image block is masked out and ignored. This mask also works well for cases where the valid information is only a small part of the image. Finally, the histogram distribution of the image is calculated according to the image mask, which will serve as the input of the global thresholding module 12 . For speed, the image can also be down-sampled and the resulting threshold applied to the original image.

Figure 3 shows the flowchart of the global thresholding algorithm in detail. This module analyzes the histogram obtained from the preprocessing module 11. First, the module 111 selects an optimal threshold T1 between the maximum value and the minimum value of the pixel gray scale, and then the modules 112 and 113 respectively select between the minimum value and T1 and Thresholds T2 and T3 are chosen between T1 and the maximum value. In a possible embodiment of the present invention, the Otsu algorithm based on the linear discriminant criterion (this is a very commonly used algorithm, source N.Otsu, "A threshold selection method from grey-level histograms," IEEE Trans. Syst., Man, Cybern., vol.SMC-1, pp.62-66, Jan.1979.) are used to determine T1, T2 and T3, that is, T1, T2 and T3 are calculated on the histogram according to the Otsu algorithm, these three The threshold satisfies T2≤T1≤T3. In module 114, if the pixel gray value in the image is less than T2, it is judged as a black pixel, which is represented as 1; if it is greater than T3, it is judged as a white pixel, represented as 0. The remaining pixels are judged as pending. It is worth mentioning that with the improvement of printing technology, there have been a large number of documents with rich background and effective text information represented by a single bright color. Therefore, in order to be able to maintain the reverse color information, the module 114 does not distinguish between the foreground and the background.

Figure 4 shows an example of global thresholding, where the abscissa is the pixel gray value, and the ordinate is the number of times each pixel gray value appears in the whole image, that is, the histogram. T1, T2 and T3 are based on the above method determines three global thresholds, where T2 and T3 are used for global thresholding.

Only by analyzing the histogram, it is impossible to determine whether the pixels falling into the interval between T2 and T3 (ie, undetermined pixels) contain useful information, so more information is needed for analysis.

Fig. 5 shows the flow chart of the local adaptive thresholding module, which is used to determine whether the pixels falling into the T2 and T3 intervals (ie, undetermined pixels) contain useful information. This module checks the pixels in the image one by one, if the current pixel is black or white, check the next pixel; if the value of the current pixel is between black and white, that is, a pixel belonging to the undetermined class, then determine a threshold for the pixel , and according to the threshold, binarize the undetermined pixel.

If the current pixel is the first undetermined pixel of the row, then module 121 uses the local feature of the current pixel to refer to the local mean and the local variance, and the method adopted is the Sauvola algorithm (see source: J.Sauvola, M.Pietkinen, " Adaptive document image binarization”, Pattern Recognition, Vol. 33, pp. 225-236, 2000.).

If the current pixel is not the first undetermined pixel in the row, add historical threshold information based on the local features, that is, the threshold determined by the last undetermined pixel. The module 122 adopts a specific method for local information and historical threshold information to determine the threshold for the current pixel, and the specific coefficient can be determined according to the application field and the characteristics of the document. For example, for OCR applications, the word extraction rate can be used as a standard to optimize the coefficients. After the threshold is selected, if the gray value of the pixel is less than the threshold, the pending pixel is binarized into black, otherwise it is binarized into white.

In a possible embodiment of the present invention, local information and historical threshold information are combined by the following formula:

T＝m*(1-k ₁ *(k ₂ *VAR+k ₃ *T _history )/R)

Among them, T is the threshold of the undetermined pixel, m is the mean value of a neighborhood centered on the undetermined pixel, VAR is the contrast of the neighborhood, T _history is the historical threshold information, k ₁ , k ₂ , k ₃ and R are all is a linear coefficient.

Document images are usually composed of characters, lines, tables, photos and charts, etc., and these different components usually have their own characteristics. But from the point of view of the binarized image, the most important information is the structure of characters, lines, tables, and internal characters. As mentioned above, noise in a binarized image can be classified into three categories: adhesion between strokes, breakage of strokes, and isolated noise points/blocks. The purpose of post-processing is to divide the cohesive ratio, connect broken strokes and remove isolated noise points, and not introduce new noise during the processing.

Figure 6 shows the flow chart of the post-processing module in detail. The basic idea is to analyze the image in an iterative manner, and whether to continue depends on the result of each iteration. First, the post-processing input is the binarized image after global and local thresholding. In each iteration, check the number of pixels with the same color in the neighborhood of each pixel. If the number is less than a certain threshold T4, the central pixel Invert the color, otherwise keep its color. The success of this method depends on the threshold and size of the neighborhood. The post-processing in the present invention selects a relatively large neighborhood, and at the same time, the neighborhood threshold is moderately increased according to the result of the previous iteration. If in a certain iteration, the number of pixels whose color is reversed is less than a certain threshold T5, it means that the noise of the image is already within a certain range, so the iteration stops. This approach effectively reduces the introduced noise.

After the processing described in Figure 2-6, an input document is converted into a binary image.

Figure 7 shows a specific example of binarization. Among them, A is the original image, B is the result after global thresholding, and C is the result after local thresholding.

The invention is not limited to the specific examples described above. It will be apparent to those skilled in the art that various combinations, changes and modifications can be made without departing from the scope of protection defined by the appended claims.

For example, a possible modification of the present invention for the preprocessing module is that the modules 103, 104 and 105 in FIG. 2 can be removed or changed. If the distribution of foreground and background pixels is relatively uniform, low-pass filtering is not required.

A possible modification of the present invention for the global thresholding module is that the global thresholding method in modules 111, 112 and 113 in FIG. 3 can be changed, such as a method based on information entropy or moment, and used to determine T1, T2 The method of T3 need not be the same either.

A possible modification of the present invention for the local thresholding module is to change the method for determining the threshold of the first undetermined pixel in the current row in FIG. 5 . Moreover, the history threshold information in FIG. 5 may be selected from the threshold of the previous pending pixel located in the same column as the current pixel. In addition, the linear coefficients used to combine local features and historical threshold information can be adjusted according to specific applications.

Claims (9)

1. one kind is carried out the image processing method of binary conversion treatment to file and picture, comprises following steps:

A) in the global threshold processing, be identified for image is carried out the global threshold of binaryzation, be divided three classes according to the pixel of described global threshold described file and picture: black, pixel white and undetermined;

B) determine an adaptive binary-state threshold for each pixel undetermined, according to described self-adaption binaryzation threshold value, with pixel binaryzation undetermined.

2. according to the image processing method of claim 1, step a) further comprises the steps:

By histogram analysis, between pixel minimum and maximum, determine first global threshold (T1);

By histogram analysis, between pixel minimum and first global threshold (T1), determine second global threshold (T2);

By histogram analysis, between second global threshold (T2) and pixel maximum, determine the 3rd global threshold (T3);

According to second global threshold (T2) and the 3rd global threshold (T3), image pixel is divided into 3 classes: pixel value is less than the black pixel that is of second global threshold (T2), pixel value is a white pixel greater than the 3rd global threshold (T3), and pixel value is a pixel undetermined between second global threshold (T2) and the 3rd global threshold (T3).

3. according to the image processing method of claim 1, it is characterized in that step b) further comprises the steps:

The employing local feature is that first pixel undetermined of every row or every row is determined described adaptive threshold;

Adopt specific mode in conjunction with local feature and historical threshold information, for each follow-up pixel undetermined is determined described adaptive threshold;

Behind the selected described adaptive threshold, if grey scale pixel value undetermined less than described adaptive threshold, then this pixel undetermined is turned to blackly by two-value, otherwise that two-value turns to is white.

4. according to the image processing method of claim 3, it is characterized in that:

Described local feature comprises image local mean value of areas and variance;

Described historical threshold information is the threshold value of current line or the previous pixel undetermined that lists.

5. according to the image processing method of claim 1, wherein, before carrying out binary conversion treatment, also comprise step:

Image is carried out preliminary treatment think that the global threshold processing provides data.

6. according to the image processing method of claim 5, wherein said pre-treatment step further comprises the steps:

File and picture is carried out low-pass filtering to remove high-frequency noise;

Determine the image masking-out according to the pixel value amplitude of variation in the image block;

If desired, can carry out down-sampled to image according to the image masking-out;

Calculate original image or down-sampled image histogram according to the image masking-out.

7. according to the image processing method of claim 6, it is characterized in that using Gaussian filter or mean filter that file and picture is carried out low-pass filtering.

8. according to the image processing method of claim 1, it is characterized in that further comprising following steps:

D) on the image of binaryzation, remove the reprocessing of noise.

9. image processing method according to Claim 8 is characterized in that step d) can further comprise following steps:

Calculate the number of pixels identical in the current pixel neighborhood with the current pixel color;

If the number of pixels that obtains is less than the 4th threshold value (T4), then with the current pixel inverse;

If reached maximum times by the pixel of inverse less than the 5th threshold value (T5) or iteration in the current iteration, iteration stopping then, otherwise recomputate the 4th threshold value (T4) and the 5th threshold value (T5), and continue iteration.

Images