Image processing method and apparatus, and server

Abstract

The embodiment of the invention discloses an image processing method and apparatus, and a server, for improving the contact ratio of a target box and a foreground target. The image processing method includes the following steps: acquiring an input box and a background frame; determining rows or columns to be cut in the target box; calculating a gradient difference matrix of the rows or columns tobe cut; and calculating the average value of the gradient difference in the gradient difference matrix, determining whether the average value is less than a gradient comparison value, and if the average value is less than the gradient comparison value, cutting the position to be cut from the target box. The image processing method takes the similarity between the input frame and the background frame as a basis for cutting the target box, calculates the average value of the gradient difference of the rows or columns in the target box of the input frame and the background frame as a comparison feature, and compares the average value with the gradient comparison value so as to determine whether the current row or column is a real foreground target, and if not, the current row or column is cutfrom the target box. Therefore, the image processing method can effectively correct the target box, and can improve the contact ratio of the target box and the foreground target.

Description

technical field

[0001] The present application relates to the field of computers, in particular to an image processing method, device and server. Background technique

[0002] In the application of image and video sequences, people are usually only interested in certain parts of them. For example, in a video surveillance system, we care about the foreground objects with significant movement, but we don’t care about the relatively static background objects, so we need The foreground and background are well distinguished, and the target frame that can accurately frame the foreground target is obtained.

[0003] In actual application scenarios, due to the influence of lighting and other reasons, the foreground object area acquired by the object detection algorithm is often larger than the actual foreground object, or some false objects are generated, which affects subsequent application scenarios. For example, in the target tracking stage, when the target is split, if the targ...

Images