Sharpness In Texture Retrieval

Texture retrieval is an important part in image processing domain. The key of texture retrieval is feature extraction. Retrieval accuracy in subsequent experiments is decided by the texture features which can precisely and comprehensively describe the picture.In this paper, sharpness is regarded as texture feature. The methodology of image sharpness evaluation is deepened and expanded in many applications, so many research results and models of method are worth considering and developing. Entropy is widely used for sharpness evaluation in image processing. The sharper the image is, the greater the entropy value of it. If the entropy reaches the max value, the image can be regarded as being of good quality. For texture image, energy is concentrated in certain area of spectrum, so it is an efficient method to measure the energy of each subband. In wavelet transform, the image quality and sharpness are closely related to the wavelet coefficients.This paper focus on the sharpness features and combines sharpness with statistical features to achieve texture retrieval scheme in multiscale domains such as DWT, DTCWT, Contourlet and PDTDFB.The main content of this paper is as follows:1, we deeply study the theory and methods of DWT, DTCWT, Contourlet and PDTDFB and compare the advantages and disadvantages of them. Then we analyze the practical value of the index FISH and FISHbb which are extracted by sharpness algorithm in texture retrieval domain.2, we discuss the improvements of sharpness algorithm in different wavelet transform. In order to adjust the architectural difference of high frequency subbands in each wavelet transform, we propose a new improved sharpness algorithm which not only suit for different wavelet transform, but also describe image sharpness precisely and comprehensively.3, we combine the improved sharpness algorithm with statistical features such as mean and standard deviation to achieve texture retrieval scheme in different multiscale domains. This method achieves high retrieval accuracy, all of this own to the good characteristics of wavelet transform and sharpness feature. The result shows that our method achieves high retrieval accuracy.