Face Feature Extraction And Retrieval Based On Wavelet Transform

Face images have different identifiable features,which can be distinguished by extracting different features.For a single individual,discriminatory features should be stable,but for different individuals,there are some differences in these characteristics,so discriminatory features must be able to distinguish different individuals.On the other hand,by reducing the dimensionality of the data,it can be classified and recognized later.Feature extraction is a very important step in face image retrieval technology.The extracted features have a direct impact on the future retrieval results.In this paper,based on dyadic wavelet transform,face image features are extracted as the main content,and three feature fusion methods are discussed in depth.In this paper,three different feature extraction methods are applied to extract and retrieve the discriminant features of face images.(1)For the color features,the HSV model which accords with the human visual characteristics is selected and quantified.The center of the set of edge points is taken as the center of the circle,and the face image space is divided into several concentric circles.By calculating the histogram of each annular region,the annular histogram can be obtained finally.Compared with the traditional histogram,the advantage of this form of histogram is that the disadvantage of spatial information does not exist.(2)For shape features,based on the method of dyadic wavelet transform to extract face contour,and using the translation invariance advantage of dyadic wavelet transform,feature extraction and retrieval of multi-feature fusion face image are realized.The edges of the image extracted in this algorithm will directly affect the retrieval performance,while the dyadic wavelet transform has translation invariance and directionality,which can effectively avoid noise.(3)For texture features,LBP operator(Local Binary Pattern,Linear Back Projection)is applied to process the texture features of the image,starting from the upper left corner of the image,in order from top to bottom,from left to right,sliding in the sliding field according to the set step size,and LBP operation is carried out on the window image obtained by each sliding.After the sliding is completed,the texture features of the image can be obtained.It shows the advantage of LBP operator in texture feature extraction.Finally,three different histograms are used to express the feature information of the image,and experiments are performed in two different standard face databases under MATLAB R2010 a environment.The results show that the extracted face feature vectors contain a variety of content information,which complement each other to describe themain content of the image.The algorithm has strong robustness to several factors such as illumination change,posture change,expression,position,time,occlusion and so on.