Research On Image Classification Of Optimized Spatial Pyramid Matching Model

With the rapid development of economy and the internet,digital cameras,smart phones and other mobile devices have entered the thousands of households.People in life use digital multimedia devices everywhere.Whether it is in learning,work,travel,or shopping and other aspects of life,we will be use mobile devices in the form of video or images to share and record their daily bit by bit.Therefore,a huge image,library is formed in the network.Then how to search and use the image from this huge image library you need becomes the focus of attention.In order to use computer to manage and organize these images,the premise is to allow the computer to analyze and understand the content of the image.Image classification is an important way to solve the problem of image understanding.And it plays an important role in the development of image retrieval technology.At present,image classification based on bag of visual words model and support vector machine has become the mainstream technology of image classification.In the bag of visual words model,the spatial information problem of the visual feature is not considered in the histogram of the visual words by using the local feature of the image.This paper introduces the spatial location information of image features by using spatial pyramid matching model.On the basis of the spatial pyramid matching model,the following improvements of the defects in the bag of visual words model are proposed:(1)Aiming at the defects of k-means clustering in constructing visual dictionary.In the third chapter,a double bag of words model is proposed to construct a visual word histogram with more ability to characterize the image.It can reduce the interference caused by the instability of the k-means algorithm and depending on the selection of the initial cluster center to experimental results.When the histogram of the visual words is constructed by using the double bag of words model,different visual words that are unstable in the cluster boundary and the very stable visual words are given different weight values,so as to obtain a visual word histogram with more characterize ability in the spatial pyramid matching model.The classification result by support vector machine shows the feasibility of the method.(2)In view of the shortcomings of the bag of visual words:1.The experimental results are affected by the instability of the k-means algorithm and dependence of the choice of the center of initial clusters.2.It does not take the visual saliency of different regions in the image into account.This paper in the fourth chapter presents an image classification method based on visual attention mechanism and spatial pyramid matching.Firstly,a fuzzy C-means clustering algorithm based on simulated annealing and genetic algorithms is used to construct the visual dictionary.Secondly,the visual attention mechanism is introduced into the spatial pyramid matching model.In real life,the different areas of the image for the human visual have different influences.Thirdly,the weighted visual word histogram is constructed by the visual attention mechanism.The image classification by using SVM shows that the method is more in line with human visual effects to judge the category of the image.