Image Classification Algorithm Using BBO_MLP And Texture Features

Image classification is an image processing method that uses different features reflected by different classes to divide the images into different classes. It quantifies the images being processed using computer instead of human to determine the class that one image or one sub-region of an image belongs to. With the development of multimedia and Internet, image classification has become a research hot spot. How to describe and classify massive images effectively thus become an urgent problem. To efficiently extract useful information from big image database, scholars put forward the characteristic vector is generated by the eigenvalues of the screening method to describe the images, which uses the texture characteristics to the research of image classification algorithm has been widely. Through the study and exploration of texture feature of intelligent optimization algorithms, this paper aiming at the existence of the biogeography-based optimization that was used to train the Multi-Layer Perceptron, often experiencing problems of premature convergence and initialization-sensitive, established a classification model based on DE_BBO_MLP and texture characteristics of the algorithm.The main research work of this paper includes the following aspects:(1) On the basis of extensive research, this paper proposed a novel Multi-Layer Perceptron training method by using hybrid differential evolution and biogeography-based optimization. Firstly, the differential evolution was introduced to the biogeography-based optimization. Then the hybrid DE_BBO algorithm was used for training MLPs to reduce these problems. In order to investigate the efficiency of DE_BBO in training MLPs, four classification datasets such as the Iris dataset, the Breast cancer dataset, the Blood Transfusion datasets and the Banknote authentication dataset were employed.(2) Image classification algorithm based on DE_BBO_MLP and texture features is established, which mainly solves the problems of the bad classification accuracy and slow convergence speed. The operation steps are as follows: firstly, three kinds of different images from the image database are selected, and the operating environment of the image classification algorithm is modeled. Secondly, texture feature is applied to represent image features, by extracting several gray image texture feature to structure corresponding the image feature vector. According to the category number which is provided by the customer and image texture feature vector to generate training samples files. Training MLP by BBO to complete the MLP define an evaluation of the habitat error fitness function and the fitness function of the global optimization, the optimized MLP is used to analyze the training sample file and get the classification model. Finally, according to the existing classification model to test the sample file information, the obtained image class number is returned to the user. The error pictures in classification are marked. Two times feedback of these pictures is conducted to improve the accuracy of image classification.(3) The results were compared with DE_BBO,BBO, PSO, GA, ACO, ES, and PBIL algorithms in a statistically significant way. The results show that the training MLP by using hybrid DE_BBO has better performance than current heuristic learning algorithms in terms of convergence speed and convergence accuracy.