Study On Content-Based Electronic Commerce Images Classification

In recent years, with the development and popularization of internet, online shopping has become a new trend so that more and more online shopping platforms appeared. With the sharply increase of commodity image data, the technical problem of how to manage a database of large-scale commodity images accurately and efficiently has been a urgently technical problem. Image classification technology based on content can classify and manage commodity effectively. According to this, we can greatly improve the efficiency and accuracy of image retrieval. We study the classification of commodity images based on the content in following three aspects:(1) Deeply discuss three visual features extraction algorithms of lower layer images, and the visual features of lower layer image include color, texture and shape. Specifically, color features include color histogram, color matrix and color coherence vector; texture features include gray level co-occurrence matrix, Gabor texture and Tamura texture, and several shape features were discussed. In addition, Study three image classification algorithms which include KNN, SVM and neural network algorithm. From theoretical and practical perspective, we analyzes and summarizes the advantages and disadvantages of the three image classification algorithms.(2) For the problem in image classification that using single classifier and the subjection function of classifier is single linear function, This paper presents a FSVM-FKNN classification algorithm based on hybrid subjection function. The algorithm builds a hybrid subjection function based on the advantage of close and linear subjection function, then apply it to integrated classifier which supports vector machine and K-nearest neighbor, Finally choose the one which has larger probability of classification as the final classification result. And we display in detail the theoretical basis and realization ideas of the proposed algorithm.(3) Present and analysis the experimental results of feature selection and classification algorithm that the feature extraction method combined color with texture can well improve the accuracy of image classification and SVM classification algorithm is the best. By comparing the FSVM-FKNN, SVM, KNN, fuzzy SVM and fuzzy KNN, the results show that the FSVM-FKNN classification algorithm based on hybrid subjection function proposed by this paper can effectively solve the problem of image classification. It improves the accuracy of image classification by 3%.