Research On Image Classification Algorithm Based On Adaboost-DBN

Image classification as an important branch of computer vision,has been generally used in many life scenes.Because the Internet develop rapidly,network data is becoming more diversified,and the complexity of image content and scale makes image classification technology facing huge challenges.Therefore,how to efficiently extract image features and design a reasonable classifier to improve classification accuracy has become a hot research topic in the field of image classification.The work for image classification in this thesis is as follows.A texture feature descriptor based on multi-resolution LNIP and a color feature descriptor based on local oppugnant color vector angle pattern are proposed.In texture feature extraction,firstly,the image is wavelet decomposed to obtain the multi-scale information of the image,and then the local neighborhood intensity pattern is used to extract the features of the multi-scale image sub-blocks;in the color feature extraction,firstly,the color space of the image is two combinations obtain vector information,and then construct local oppugnant color vector angles to extract color features.Simulation experiments were conducted on the Corel-1K data set,and the classification results of the two proposed feature descriptors in the DBN classifier reached83.4% and 83.6%.Compared with other feature descriptor classification results promote under the same classifier.As an ensemble algorithm,Adaboost can form a strong classification model with multiple homogeneous classifiers to improve classification accuracy.In the design of the classifier algorithm,this thesis improves on the shortcomings of the Adaboost algorithm that requires a high classification effect of the base classifier,and uses the DBN as the base classifier to design the single-input Adaboost-DBN classification algorithm.The input of the algorithm is the concatenation of texture features and color features.In addition,this thesis improves the structure of the traditional Adaboost algorithm,and resets the sample weight update formula and strong classifier output formula,and proposes a two-input Adaboost-DBN classification algorithm.The texture features and the color features are input into the algorithm in parallel fortraining,and the misclassified samples are adjusted through mutual feedback between the output results of the two features to improve the learning ability of the algorithm.Experiments were performed on the Corel-1K data set,and the classification results of the two proposed algorithms reached 84% and 85.2%.The classification results of this algorithm and other algorithms are compared,and experiments prove that the proposed algorithm is better.