Speckle Processing For OCT Image Based On Bayesian Least Mean Square Error Algorithm

Optical Coherence Tomography is a new type of imaging technology which hasseen rapid development in the past decade. It integrates the coherence feature ofinfrared ray with heterodyne detection to create images for shallow surfaces ofbiological tissues. With high resolution, sensitivity and non-destructive testing, OCThas gained a wide application in the field of biomedicine.It is inevitable that in OCT system, speckle noise is generated during the processof imaging which could reduce image resolution and contrast, and consequentlyaffect detection of subtle structure in biomedical tissue. In this paper, a detailedanalysis of the speckle noise suppression algorithms available has been made and anew algorithm based on non-linear logarithmic space for Bayesian minimum meansquare error estimation against their shortcomings on suppressing speckle noise ispresented. The novelty of the algorithm stems from conditional posterior samplingmethod based on the statistical characteristics of speckle noise used to take samplesin logarithmic space, estimation of posteriori distribution of noise-free data by thenon-parametric estimation method and use of the Bayesian minimum mean-squareto estimate noise-free data. The experimental results indicates that the algorithmcould noticeably improve the SNR and ENL of images obviously compared with thetraditional wavelet transform algorithm and median filtering algorithm, but it is notperfectly sufficient in keeping the edge. According to the shortcomings of thealgorithm in term of edge retention, the weight of image pixel is adjusted bycombining image pixel space close degree with image pixel gray similarity level asweight of image pixel. The experimental results show that the improved algorithmcould suppress the speckle noise more effectively while preserving edge feature ofimages than the Bayesian minimum mean square error estimation and the medianfiltering.