A large-scale image sub-block retrieval method based on a deep Hash network and sub-block reordering

Abstract

The invention discloses a large-scale image sub-block retrieval method based on a deep Hash network and sub-block reordering. The large-scale image sub-block retrieval method comprises the following steps: preparing a deep network weight pre-trained by a large-scale image database and a texture picture library to be trained; extracting picture pairs from the texture picture library, when the two pictures have the same label, determining the two pictures as positive samples, and taking the two pictures with different labels as negative samples; inputting the picture pairs into two same shared weight networks for training in pairs, and setting a loss function to binarize a network output result; quickly obtaining similar pictures by using the network as a Hash code extractor; after subblockreordering is conducted on the similar pictures, quickly mapping subblocks in an original picture and obtaining a large number of similar sub-blocks. According to the method, the calculated amount isgreatly reduced, rapid retrieval of large-scale image sub-blocks is achieved, and the method can be applied to various image enhancement methods.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a large-scale image sub-block retrieval method based on a deep hash network and sub-block reordering. Background technique [0002] Pictures fill every aspect of modern life. With the advancement of network technology and the popularization of the Internet, images, a communication medium containing a large amount of information, are showing an explosive growth trend. People are no longer satisfied with simply reading text or listening to sound. Pictures can give richer information and more intuitive feelings, and can also give a more subjective experience to the dissemination of information. However, although the image medium has many advantages, it is more prone to distortion due to its high requirements for transmission. When the network situation fluctuates, the transmitted pictures often have low resolution or cover noise, which affects the experience. Bad influence...

AI Technical Summary

Problems solved by technology

[0004] Facing the problem of large-scale image sub-block retrieval, the first thing to solve is the data scale. For a picture with a size of 512*512, if the image sub-blocks are densely sampled according to the size of 7*7, then 256036 image sub-blocks will be obtained. The amount of calculation for finding approximate sub-blocks in a single image is acceptable, but it is completely unacceptable to expand to the scale of multiple images or even databases

Existing image retrieval methods are mainly aimed at the image level, and rarely involve retrieval at the image sub-block level. Due to the small scale of the image sub-block itself, it is difficult to contain high-level information

If you simply adopt the method of intensive cutting and brute force to find the nearest neighbor, the efficiency is too low; the commonly used nearest neighbor method, such as the famous k-d tree, in the face of such a large-scale data, the efficiency will drop sharply; Small image sub-blocks are difficult to improve, and they are not suitable for practical applications

Images