Image retrieval method based on CNN feature vocabulary tree

Abstract

The invention discloses an image retrieval method based on a CNN feature vocabulary tree, and aims to solve the technical problem of low accuracy existing in the existing vocabulary tree method. The implementation steps are as follows: first generate the derivative images of each image in the image library, and extract the CNN features of each image in the image library, then construct a CNN feature vocabulary tree according to the extracted CNN features, then generate derivative images of each image to be retrieved, and extract The CNN feature of each image to be retrieved, by comparing the path of the CNN feature of each image to be retrieved and its related image in the CNN feature vocabulary tree, calculate the distance between the image to be retrieved and its related image, and compare the image to be retrieved and its related image The distance is combined with the initial similarity, and finally the retrieval results of each image to be retrieved are output according to the comprehensive similarity of each image to be retrieved and its related images. The image retrieval accuracy rate of the invention is high, and can be used in medical image computer aided diagnosis systems and image search systems.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to an image retrieval method, in particular to an image retrieval method based on a CNN feature vocabulary tree. It can be used in medical image computer aided diagnosis system and image search system. Background technique [0002] Image retrieval is the process of comparing the similarity between the description content input by the user and the images in the image library according to a certain similarity comparison mechanism, and returning similar images. With the development of science and technology, the number of images is increasing explosively, and the difficulty of image retrieval is also increasing. [0003] According to the different input description content, image retrieval can be divided into text-based image retrieval and content-based image retrieval. Text-based image retrieval requires a lot of manual labeling, which consumes a lot of manpower and material ...

AI Technical Summary

Problems solved by technology

This method uses the feedback information of user search results to modify the semantic mapping, which improves the accuracy of image retrieval. However, due to the lack of effective user feedback information in practical applications, the actual accuracy of this method is limited. At the same time, the SIFT feature used is a Artificially designed local image descriptors cannot fully describe the high-level semantics of images, and the retrieval accuracy for images with complex content is low

Images