Panoramic-image-established vision case searching method

Abstract

The invention provides a panoramic-image-established vision case searching method. The panoramic-image-established vision case searching method mainly includes the steps of convolutional neural network (CNN) descriptor searching, representative vector set searching, data set image aggregating, image aggregating and querying and panoramic-image matching, wherein specific descriptor expression suitable for case searching is obtained through the CNN, sum storage vectors are defined as summation vectors, pseudo-inverse vectors are defined as incident vectors, the computing cost of searching-time similar querying is reduced by aggregating data set images, the searching quality is improved in the mode that querying vectors are aggregated with the incident vectors, and a panoramic image is established with the implicit method and the explicit method. According to the method, the speed is increased by aggregating the data set images, memory is expanded, and meanwhile performance losses are reduced; space is greatly saved by aggregating descriptions of all the images in each group, and meanwhile the computing cost of searching-time similar querying is reduced.

Description

technical field [0001] The invention relates to the field of visual instance search, in particular to a visual instance retrieval method constructed using panorama images. Background technique [0002] With the rapid development of multimedia and network technology, the application of image and video information is becoming more and more extensive. Effective management of increasingly large image and video databases and visual information has become an urgent problem that people need to solve. In recent years, deep learning has made major breakthroughs in various computer vision tasks. One of the important factors is its powerful nonlinear representation ability, which can understand the deeper information of images. Among them, visual instance retrieval is an important research direction, which can find those pictures that contain the same object as the query picture from the candidate picture library, which can be applied to product search, vehicle search and image-based g...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of taking up too much memory, long query time, and high computing cost, the purpose of the present invention is to provide a visual instance retrieval method constructed by using a panorama, and first obtain a specific descriptor representation suitable for instance search through a convolutional neural network , define the sum storage vector as the sum vector, and the pseudo-inverse storage vector as the association vector, use the aggregated data set image to reduce the computational cost of query time similarity search, and use the association vector to aggregate the query vector to improve the search quality, through both implicit and explicit Method to build a panorama

Images