Object-level depth feature aggregation method for image retrieval

Abstract

The invention relates to the field of digit media and provides an object-level depth feature aggregation method for image retrieval. First, an unsupervised method is used to generate candidate regions that may contain objects; then corresponding convolution neural network characteristics are extracted; finally, the area features are aggregated to obtain image feature representation having high robustness for image transformation for the use of image retrieval applications. The present invention addresses the lack of geometric transformation and spatial layout invariance of existing models, and the object-based mode is adopted to solve the problems in the prior art; the image features generated by the method have high robustness on image geometric transformation and spatial arrangement transformation; the accuracy of image retrieval is increased; the obtain image is quit compact and concise so that complexity of similarity calculation among images is reduced and retrieval efficiency is increased.

Description

technical field [0001] The invention belongs to the field of digital media, and relates to an image retrieval-oriented object-level deep feature aggregation method. Background technique [0002] Content-based image retrieval, as an important research problem in the field of computer vision, has received extensive attention from domestic and foreign scholars in the past decade. Content-based image retrieval refers to finding images similar to the query image from the image database. Due to different factors such as angles, distances, and environments when shooting, similar or identical subjects will have great changes in different images, such as changes in scale, perspective, and layout. Therefore, generating an image feature that is highly robust to various image variations is the key to solving image retrieval problems. [0003] Compared with traditional image features based on artificial design, learning-based methods, especially convolutional neural networks, have show...

AI Technical Summary

Problems solved by technology

Although these methods take into account the local information of the image, making the feature more robust to various changes than the global method, there are still some defects in these methods.

For example, use the sliding window method to obtain the image area (refer to the article "Multi-scale orderless pooling of deep convolutional activation features" published by Yunchao Gong, Liwei Wang, Ruiqi Guo, and Svetlana Lazebnik on pages 392-407 of the European Conference on Computer Vision 2014 ), because the color, texture, edge and other visual content of the image are not considered, a large number of regions with no semantic meaning are generated, which brings redundancy and noise information to the subsequent aggregation process

In addition, the maximum pooling algorithm commonly used in regional feature fusion (refer to the article "Objectlevel deep feature pooling for compact image representation"), because only the maximum response of the feature is retained without considering the correlation between features, a large amount of information is lost, and the discrimination of the final image features obtained is reduced

Images