Image Classification Based On Comprehensive Feature

Classification as the basis of image recognition, has become an important tool for people to analyze and understand the image information. Rapid and efficient classification has been increasingly demanded with the arrival of big data era. It's no longer able to cope with the challenge of the era of big data to use the traditional manual annotation method. How to use the image content effectively and extract useful information from the massive image database has become a challenging research topic. From 1990 s content-based image classification technology has drawn wide attention from all walks of life.Content-based image classification system is to describe the image content mainly by extracting the underlying characteristics of the image and realize image classification combined with machine learning methods. Main accomplished work is as below:Firstly, a new way of non equal interval quantization is proposed to quantitate the color value based on HSV space. The image color feature extracted by this method has the advantages of low feature vector dimension and low light sensitivity with the classification accuracy reached 65.6%. The classification accuracy is comparable to the traditional 72 level quantization method, but the speed of feature extraction and classification training is faster.Secondly, a method of image features extraction is proposed in this thesis. Herein by combining the image color and texture features as the image's comprehensive feature for image classification. The method can describe the image content more comprehensively and avoid the limitation of single feature. The simulation results show that the classification accuracy can be increased by using the comprehensive feature classification method compared with the single feature classification method.Thirdly, a comprehensive image feature weighting algorithm is presented in this paper. A single feature with particularly low classification accuracy can reduce the classification accuracy of comprehensive feature. To solve this problem, firstly we should find the single feature with particularly low classification accuracy and then reduce the effect of the single feature on the classification results by the weighting method. The average classification accuracy is also improved to a certain extent by this weighting algorithm.