Image concept detection method based on Hamming embedding kernel, and Hamming embedding kernel thereof

Abstract

The invention discloses an image concept detection method based on a Hamming embedding kernel. The image concept detection method comprises the following steps: extracting partial interest points of a training image, and establishing a visual dictionary through SIFT (Scale-Invariant Feature Transformation); generating binary signatures of the partial interest points, and performing off-line training on the Hamming embedding kernel; generating binary signatures of the partial interest points; calculating the Hamming distance between the binary signatures; calculating the Euclidean distance between random partial interest points; looking for the optimal matching of the partial interest points; generating the Hamming embedding kernel; performing image concept detection by utilizing the Hamming kernel. According to the image concept detection method disclosed by the invention, the Hamming embedding is improved and merged into the inner kernel of an SVM (support vector machine) for image concept detection, so that the capability of distinguishing contents and concepts of different images, of a bag model of visual words, is enhanced. The invention further discloses the Hamming embedding kernel.

Description

technical field [0001] The invention belongs to the technical field of image concept detection, and in particular relates to an image concept detection method based on a Hamming embedding kernel and the Hamming embedding kernel thereof. Background technique [0002] Concept detection, as the most basic step in content-based image retrieval, has been intensively studied in the past few years, and many effective methods and features have been proposed. For classifiers, SVM (Support Vector Machine) has been widely used. To improve the performance of SVM, a method that can accurately describe the distance between image samples plays a key role. [0003] Among various features to represent images, Bag-of-Visual-Words (BoW) features have achieved great success due to their high efficiency and ability to distinguish different concepts using local image information. In a typical BoW feature extraction process, the local interest points (also called keypoints) of the image are firs...

AI Technical Summary

Problems solved by technology

By ignoring these differences, the BoW method largely loses the discriminative power of SIFT descriptors, resulting in a significant decrease in the accuracy of distance estimation between different image samples, which ultimately affects the performance of the classifier.

Images