Scene Image Classification Method Based on Ring Space Pyramid and Multi-kernel Learning

Abstract

The invention discloses a scene image classification method based on an annular space pyramid model and multi-core learning, including: establishing a training image set and a test image set; a multi-feature extraction stage, including extracting Dense-SIFT features, L-Gist features and color Color features; using the secondary K-means++ clustering to train the dictionary, the process of secondary clustering is performed for each feature extracted, and then the second clustering process is performed on the set of visual dictionaries generated by the first clustering Clustering to obtain the total visual dictionary; image feature encoding stage, by dividing the image into an annular space pyramid, for each sub-image block after the pyramid division, a vector representation is formed based on the visual dictionary; in the multi-core learning stage, an annular space pyramid is used Divide the image, assign a kernel function to each sub-image block, and assign a kernel function to the color feature; classification and judgment stage. The present invention uses a complementary combination of Dense-SIFT features, L-Gist features and HSV global color features to represent scene images, which can more effectively represent the complete information of images than conventional single-feature methods, and can better realize scene classification.

Description

technical field [0001] The invention belongs to the field of machine learning and digital image processing, in particular to a method for classifying scene images based on annular space pyramid and multi-core learning Background technique [0002] In recent years, due to the rapid development of multimedia and Internet technology, the rapid expansion of image information resources has been greatly promoted. While massive image resources bring great convenience to our work and life, how to manage and quickly retrieve our Images of interest are becoming more and more difficult. Therefore, in the face of vast image resources, relying on the traditional manual labeling method is not only time-consuming and laborious, but also has subjective uncertainty, which obviously does not meet the needs of the rapid development of today's multimedia information age. Then, how to use smart devices such as computers to complete the automatic classification and efficient management of image ...

AI Technical Summary

Problems solved by technology

[0005] In view of the above technical problems, the present invention proposes a scene image classification method based on annular space pyramid and multi-core learning, extracts local feature Dense-SIFT and local Gist feature L-Gist from the scene image, and combines the global color feature of HSV color space To combine and represent scene images, it overcomes the problem of information loss caused by using a single feature to represent images in traditional classification methods; uses the encoding method of three-level spatial pyramid aggregation to encode these features; in order to increase the spatial information when classifying scene images and each image Blocks have different contributions during classification, and the method of dividing and weighting the ring space pyramid is used to increase the spatial information between the scene image features; Each image patch of is assigned a kernel function, and by learning the weights of each kernel, the synthetic kernel with the strongest distinguishing ability is obtained

Images