Image Classification Application Based On Optimization Of Support Vector Machine

Image classification is applied to image automatic recognition todifferent semantic category or pre-set different image categories. With themature application of multimedia technology, image automatic classificationhas become a key task in application fields such as image retrieval and videoclassification, network information filtering, RS system positioning, visualscene target tracking applications, etc. At the same time, many classificationand image representation technology have obtained the considerabledevelopment.After artificial neural networks, Support Vector Machine have becomeanother very effective and popular technique in the field of machine learning,and widely applied in the image recognition and classification.At present, to realize image classification with Support Vector Machine,it is supposed to accomplish digital image feature extraction in advance. It’smore frequently for color features, textural features, shape features andspatial relationship features to be used among the features of image. In orderto close the distance between the high-level semantic and bottom semantic, itrequires pre-processing such as image feature organizations to use forclassifier. Meanwhile to improve the image classification effect by the Support Vector Machine classifier, the method of how to optimally choosekernel functions and kernels’ parameters would be researched.This paper first analyzes the domestic and foreign related subjects’research dynamic; Secondly recommend the image classification realizationprocess with Support Vector Machine briefly, study all kinds of imagefeatures and the feature extraction method widely used now at great length;Again through the research of kernel functions’ algorithm and kernels’parameters selection method commonly used in SVM classifier, optimallychoose SVM classifier kernel functions and kernels’ parameters method inorder to strengthen the classifier performance, at the same time use efficientimage feature extraction and organization methods match with SVM toovercome the weakness of less fusion prior knowledge from traditionalkernel functions; Then use the standard image database Caltech-101,Caltech-256as classification sample on the Matlab platform to test abovearithmetic and comparative analysis, verify the effectiveness of theoptimization method; Finally, summarize the main work of this paper and theSVM and look into the distance of the application of image classification andthe SVM method.