Image binarization method, device and video analysis system

Abstract

The invention relates to an image binarization method, a device and a video analysis system. The method comprises steps: binarization processing is carried out on a to-be-processed area or an image, and a binarization processing threshold is acquired; a neural network is adopted for training pixels whose pixel values are not within a threshold shift range in the to-be-processed area or the image, and binarization is carried out on a training result, wherein the lower limit of the threshold shift range is the threshold minus a predetermined shift value, and the upper limit of the threshold shift range is the threshold plus the predetermined shift value. According to the image binarization method of the invention, basic binarization processing is firstly carried out, neural network algorithm is adopted for training pixels whose pixel values are obviously higher than or lower than the threshold used for basic binarization processing (the pixel values are not within the threshold shift range), self-learning features of the neural network are used, and binarization of the image is more intelligent.

Description

technical field [0001] The invention relates to the technical field of video intelligent analysis, in particular to an image binarization method and device, and a video analysis system. Background technique [0002] In digital image processing, binary image occupies a very important position. The binarization of the image is beneficial to the further processing of the image, which makes the image simple, reduces the amount of data, and can highlight the outline of the target of interest. Many related technologies must use binarized images, such as character recognition, document image analysis, target detection, etc. [0003] Various algorithms have been developed to binarize images, such as the maximum inter-class variance method (also known as Otsu algorithm, OTSU for short), Bernsen algorithm (a typical local threshold method), etc. However, these algorithms have certain limitations, because the principles they are based on are not necessarily suitable for all different...

AI Technical Summary

Problems solved by technology

However, each of these algorithms has certain limitations, because the principles they are based on are not necessarily suitable for all different complex situations, such as low contrast between the image target and the background, whitening, and uneven illumination due to environmental changes.

Taking OTSU as an example, it only produces a good segmentation effect on images with a unimodal variance between classes. When the size ratio of the target and the background is very different, the variance criterion function between classes may appear bimodal or multimodal, and it is difficult to obtain satisfactory results. split effect

Images