ELM and DE combined image classification method based on multichannel features

Abstract

The invention provides an ELM and DE combined image classification method based on multichannel features, and the method is characterized in that the method comprises a training process and a prediction process; the training process comprises the steps: a1, taking faces of persons with different gender at different ages in different light illumination scenes as positive samples, and taking small blocks of other body parts except the faces as negative samples; a2, carrying out the unified conversion of the sizes of the samples to a specified size, and carrying out the Gaussian filtering; a3, extracting the integral channel features of the above sample images, which comprise the gray color channel features, gradient direction histogram channel features and gradient amplitude channel features; a4, enabling the integral channel features extracted at step a3 as the input of ELM, carrying out the image classification and training, carrying out the improvement and amplification of the ELM through a DE differential evolutionary algorithm, enabling the classification effect of the ELM to be optimal, and obtaining a trained classifier; the prediction process employs the trained classifier for the classification of the images.

Description

technical field [0001] The invention relates to the field of image classification, in particular to an image classification method based on the combination of ELM (extreme learning machine) and DE (differential evolution algorithm) based on multi-channel features. Background technique [0002] Image classification is a fundamental problem in many important research fields in computer vision and image processing today. Good image classification technology can effectively solve problems in other scientific research fields, such as image retrieval, remote sensing images, 3D reconstruction, and so on. The purpose of classification is to establish a classifier based on existing image features, which can predict unknown image types. For example, land use maps, vegetation coverage maps, and other maps are obtained through classification of remote sensing images, and these maps are used as the basic maps for the next step to carry out environmental and land use; in medicine, X-ray ...

AI Technical Summary

Problems solved by technology

However, these methods have different degrees of problems

First of all, the texture, shape and color space features used are not enough to express the characteristics of the image

Secondly, the SVM classifier needs feedback to adjust parameters, which takes a long time and cannot achieve fast classification results.

Images