Research On Noise Degradation And Contour Extraction Algorithm For Eye Fundus Oct Images

The continuous development of detection and imaging technology produced more and more medical imaging methods, one of which is optical coherence tomography(OCT). Ophthalmic imaging is a representative field that applies OCT technique. The OCT devices designed to check for eye diseases have been widely used. In order to improve the automation level of ophthalmic OCT systems, related processing software has become a necessary part in it.We will have a research and discussion on the key processing algorithm of eye fundus OCT images. They mainly include speckle degradation and contour extraction for target images.After analysis on the characteristics of eye fundus images and the abilities of some classical image restoration methods, we proposed in the thesis a speckle suppression algorithm based on wavelet decomposition and reconstruction. Using the correlation of wavelet coefficients in the space domain, the algorithm gets its suppression factors by calculating the probability if the processing wavelet coefficient is a"signal coefficient". Besides, we have designed three suppression strategies for images having different noise characteristics, and then the algorithm can be used to degrade noise energy and reserve signal energy in the target OCT image. Comparative experiments show that our proposed algorithm has better processing results for eye fundus OCT images, and can improve the SNR and contrast of it more effectively.Another important work of the thesis is contour extraction for eye fundus OCT images. After analysis on the contour characteristics of the target images and the performances of some classical contour extraction methods, we designed a contour extraction algorithm for eye fundus OCT images based on multi-scale active contour model. In the algorithm, we get initial contour of the evolution curve using directional features of the Sobel gradient operator and statistics refinement method. Then we designed the energy function for Snake according to the contour features of eye fundus OCT images, the innovation lies in that a distance repulsion force is added into the energy function in order to avoid the overlapping or intersection of the two evolution curves. The ground truth is derived by artificial mark. Results show that the algorithm is effective for most images, and it has strong robustness for noises in the images and"glitches"in the contours. The detection accuracy has reached the requirements of clinical applications.