A Snowflake Noise Detection Method Based on Information Entropy

Abstract

The invention relates to a snowflake noise detection method based on information entropy. The present invention is based on obtaining the overall distribution state of gray value of pixels, by transforming the front and rear image frames into gray space respectively, and then clustering the gray values ​​of all pixels into n intervals, in the form of gray histogram Show its distribution state, and calculate the respective information entropy, and finally judge whether there is snowflake noise in the video stream by comparing the change of the information entropy of the frame before and after. In the present invention, when the actual calculated value is not less than the established threshold and the value of the next frame is greater than the previous frame, or when the actual calculated value is less than the established threshold but the previous frame is marked as a noise frame, we finally determine that there is real snowflake noise , otherwise it is considered a normal image. The present invention has higher accuracy in detecting video snowflake noise, and also makes a good foundation for obtaining better visual effect after removing snowflake noise.

Description

technical field [0001] The present invention relates to a method for detecting snowflake noise in a video stream, in particular to a method for detecting snowflake noise by using the characteristic value of image information entropy as an evaluation index, performing corresponding comparison processing on frames before and after the video stream, and obtaining the final noise detection result method. Background technique [0002] With the rapid development of computer technology and communication technology and other fields, people's requirements for the visual effect of image data are also increasing, but the noise caused by signal interference and line problems is still one of the main factors affecting the visual viewing effect. including snowflake noise. Now, more and more attention has been paid to the processing in this aspect, and detecting noise is a key step in removing noise. [0003] So far, there have been some viewpoints and methods for the detection of snowfl...

AI Technical Summary

Problems solved by technology

In Qiu Mingjie's paper "Research and Implementation of Surveillance Video Image Quality Diagnosis Method", it is proposed to use noise points and SNR value to comprehensively judge noise, but it does not take into account the characteristics of snowflake noise belonging to the global distribution of images, resulting in some unnecessary false detections ; In the journal article "Image Quality Detection in Video Surveillance" by Liu Qu, Zhang Guimao, and Liu Xiang, a method of using small windows to calculate the variance of local images is proposed, which is also based on the gray value of noise points and normal pixels difference, but it is based on the detection of the image itself rather than the detection of the video; in the patent "A System and Method for Monitoring Snowflake Noise in Surveillance Video" (Patent No.: 201410636977.2), by randomly selecting several images of the same frame before and after Sub-blocks with the same size, compare their mean square error for preliminary judgment, and then calculate the SNR value to finally judge the noise. This method starts from the block, although it reduces the workload to a certain extent than starting from the pixel point, but there are chances. False detection

In the patent "A Video Anomaly Detection Method Based on Machine Learning" (Patent No.: 201310722563.7), it is mentioned that the noise points are detected by global detection and block detection in two steps to determine whether there is snowflake noise in the video stream. It first uses pixels The gray value changes of the frames before and after the video stream are used to preliminarily determine the noise points, and then the spatial layout characteristics of the snowflake noise points in the image are used as the detection basis for the existence of snowflake noise. This method has a certain degree of credibility in theory. , but after a large number of experimental verifications, the detection effect obtained based on these two steps is not ideal. Excluding the errors generated by the algorithm itself, such a theory still cannot handle many situations. The analysis is as follows: the difference operation can detect Whether there are a large number of irregular motion noises in the image, but the algorithm has certain limitations when applied to surveillance video: in rainy and snowy weather, it is sometimes difficult to distinguish noise and weather interference factors due to the capture of small raindrops in the video screen; When the area occupied by moving objects in the picture is too large, it is easy to judge the image without snowflake interference as the image with interference, which leads to unsatisfactory detection results

Images