EMD (empirical mode decomposition)-based intelligent document image block detection method

Abstract

The invention discloses an EMD (empirical mode decomposition)-based intelligent document image block detection method. The method includes: subjecting a document image to grey level transformation; calculating an average, a variance and a horizontal projective histogram of a document gray level image; subjecting the histogram to EMD decomposition, and taking a summation result of a last eigenmode modular function component and a residual error as a trend line of the histogram; calculating x-coordinates of intersections between the trend line and the histogram, and calculating spacing distances of the x-coordinates; carrying out image subblock segmentation according to the spacing distances, and calculating an average and a variance of each image subblock; judging whether the image subblocks are background or not according to the average and the variance of the document gray level image; subjecting the histogram to EMD decomposition, and taking a summation result of a last and the averages and the variances of the image subblocks, and processing correspondingly according to judgment results. The method is not only applicable to file images degraded by complicated imaging environments but small in calculated amount and applicable to portable devices. In addition, the method is widely applicable to the field of file image detection.

Description

technical field [0001] The invention relates to document image detection and processing technology, in particular to an EMD-based intelligent document image block detection method, which is suitable for adaptive threshold segmentation of degraded document images. Background technique [0002] technical term explanation [0003] EMD: Empirical Mode Decomposition, the English full name is Empirical Mode Decomposition. [0004] It has great application value for distinguishing text from the background of the document image, that is, converting the grayscale image of the document image into a binary image, and in many applications, it is mainly used in optical character recognition, automatic bank check processing, Signature verification, etc. At present, many algorithms have been proposed to distinguish text from the background of document images, but due to the limitation of imaging environment, document images are often degraded by factors such as complex backgrounds, non-u...

AI Technical Summary

Problems solved by technology

At present, many algorithms have been proposed to distinguish text from the background of document images, but due to the limitation of imaging environment, document images are often degraded by factors such as complex backgrounds, non-uniform intensity, shadows, etc. in the text, it becomes very difficult

[0005] In order to solve these problems, many binarization algorithms have been proposed. These algorithms are mainly classified into two categories, which include: 1. Global threshold segmentation algorithms, such as Tsai, Johannsan, Kapur and other algorithms. However, these methods are not applicable For source images containing various background patterns or uneven backgrounds; 2. Local threshold segmentation algorithms, such

Images