Image Quality Assessment And Computer-aided Detection Based On Feature Extraction

With the rapid development of digital terminals and the Internet, digital images and medical image processing technology have been rapidly developed. More specifically, the image feature extraction as a key technology has been paid more and more attention. The quality of image feature extraction largely determines the performance of the algorithms pros and cons.In the field of natural image processing, there will be data loss and the introduction of an external signal during the process of digital image acquisition, compression, storage, and transmission process, resulting in digital image distortion. The distortion makes it inconvenient for people to understand and use the photo. In order to improve the acquisition, compression, storage and transmission technology affecting image quality, we often need to have feedback the evaluation of image quality indicators, so the importance of image quality assessment are also increasingly prominent for the improvement of image processing methods. In the medical field imaging, the radiologist confirms the lesions of the image, which is an important work. But the accuracy of such judgments often depends on doctor’s clinical accumulated experience, making it difficult to popularize. Because the early lesions are difficult to observe with the naked eye, which is the inconvenience of early diagnosis. The early diagnosis is of great significance, since the results of early diagnosis can be used for confirmation of lesion. Treating the lesion area can prevent the development from being benign to malignant. Especially in the field of early diagnosis of breast cancer, a large number of computer-aided detection technologies are proposed and applied to assist radiologists diagnose early breast cancer. Early breast cancer lesion area which exists in the form of presentation is different from the normal tissues in medical images. The breast cancer lesion area can be distinguished between features extracted from lesion area and features extracted from the non-lesion area. The classification is available for radiologists through image processing technology with rich and direct image information to help doctors make the final diagnosis, which results in improving efficiency and accuracy of diagnosis.This thesis presents a research based on the non-reference image quality assessment and computer-aided diagnosis feature extraction. The extraction of features are chosen for specific characteristics as follows:(1) A no-reference image quality assessment method using topology independent component analysis for feature extraction. Taking into account of the natural scene statistics including a non-Gaussian characteristic, topology independent component analysis can be employed for the feature extraction of this non-Gaussian characteristic. The common feature can be obtained by the training of universal features, which can be compared with the features extracted from distorted images. Then the difference is quantified and mapped to the human eye objective image quality assessment scores. Experimental results show that the algorithm has a good consistency with subjective perception of the human eye. The algorithm is compared with the classical algorithms, which shows that the proposed algorithm has a good performance overall.(2) A computer-aided detection of breast calcifications based on fractal for texture feature extraction. Breast calcification is an effective symbol for early diagnosis of breast cancer. It has a clear distinction between non-calcified tissue texture and calcified tissue texture. As the fractal theory is often used to extract image texture feature, this thesis employs fractal theory for extracting texture features of breast calcifications. Then with the classifier learning method, the texture feature can be classified into two groups, thereby being used to judging whether the tissue contains breast calcifications.