Image feature extraction method based on the combination of src‑dp and lda

Abstract

The invention discloses an image feature extraction method based on the combination of SRC-DP and LDA, which mainly solves the problem that the existing feature extraction method cannot more accurately describe the sample information because only the reconstruction relationship or the discrimination relationship is considered, so that the image classification results are not accurate. Ideal question. The implementation steps are: 1. Input the training samples, calculate the intra-class and inter-class dispersion matrix of the samples, initialize the projection matrix, 2. Project the training samples, and sequentially solve the sparse representation coefficients of the projected samples; 3. Calculate the projections respectively Intra-class and inter-class reconstruction of the dispersion matrix of samples; 4. Construct the objective function to solve the new projection matrix; 5. Iterate steps 2-4 until the number of cycles is greater than the given initial value, and output the final projection matrix. The invention enhances the accuracy of image classification, improves the classification and recognition rate, and can be used for identifying people's identities in police systems or searching for objects in image shooting.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a feature extraction method in image recognition, which can be used for identification of persons in police systems or search for objects in fields such as image shooting. Background technique [0002] Image recognition is one of the hot and challenging research directions in machine learning, pattern recognition and computer vision. Image recognition technology has been widely used in intelligent transportation, public security, biomedicine, e-commerce, remote sensing technology, military and multimedia network communications and other fields due to its advantages of simplicity, efficiency, safety, and low cost. Images are often affected by imaging factors such as viewing angle, illumination, and occlusion, which brings great challenges to image classification. Image recognition belongs to the problem of pattern recognition in high-dimensional space. Therefore, when r...

AI Technical Summary

Problems solved by technology

[0004] One is that only the local reconstruction of the data is considered, and the global discriminant information of the data is ignored, which makes the overall classification result unsatisfactory.

[0005] The second is that because only the reconstruction residual is considered, the discriminant structure cannot be better described, especially in the case of a large number of categories, the SRC-DP method cannot be effectively identified

For example, for the c classification problem, it can find a c-1 projection direction, thereby compressing the dimension to c-1. Therefore, although this method has outstanding data compression capabilities, it can effectively reduce the amount of data for subsequent processing, and can effectively However, when using this method for image recognition, only the overall discriminant information of the sample is considered, and the sample is not analyzed from the perspective of reconstruction, which leads to unsatisfactory classification results.

Images