A Flower Image Classification Based On Cellular Automata And Weighted Feature

With the improvement of life quality,more and more people have become growing flowers to cultivate temperament.Faced with such a wide variety and mixed colors flowers,it has great values both on theory and practice to research on an effective flower image classification method because it can help people better understand the general morphology and the habit of the flower,and accordingly improve the culture level of flower.Image segmentation and feature fusion are two main steps to improve the accuracy of flower image classification.However,the traditional image segmentation method often results in poor segmentation due to the complexity of the flower image background.The general feature fusion method simply splices multiple features together,and the differences of different features' contributions to flower images classification are not taken into account,thus affecting the effect of classification.To further improve the classification efficiency of flower images,a flower image classification method based on cellular automata and weighted feature fusion is presented in this thesis.The mainly research work includes the following three aspects:(1)A method of extracting flower body region based on cellular automata is presented.Firstly,the image preprocessing is performed on the flower image.On this basis,the SLIC algorithm is used to divide into N small pixels.The color space and the distance space are compared with the marginal seed of the classification to obtain a saliency map from the background.Secondly,by using a new transmission mechanism of cellular automata,an optimized saliency map is generated according to the corresponding rules.Then,the operation of converting the gray-scale map into a binary image on the saliency map is completed according to an appropriate threshold,which is decided by the maximum interclass variance method.Finally,on the basis of the original flower image,the white part of the binary image is filled,and the flower body area is obtained.Experiments results on the 17-flower datasets show that this method iseffective.(2)A weighted feature fusion method based on flower body region is proposed.The general feature fusion method simply splices multiple features together,and the differences of contributions of different features to flower images classification are not taken into account,thus affecting the accuracy of the flower images classification.In order to improve the accuracy of flower image classification,the color features and the local features of the flower body area,which is obtained from the above method(1),are firstly weighted and fused.Then the flower images are classified by using SVM.Finally,the method of flower image classification is verified by experiments on 13-flower and102-flower image datasets.(3)A prototype system of flower image classification based on cellular automata and weighted feature fusion is developed.By using Matlab7.0 and VB6.0 as the development tools,a prototype system of flower image classification based on cellular automata and weighted feature fusion is designed and implemented.