For any coherent imaging systems including ultrasound, synthetic aperture radar and optical laser, the multiplicative speckle noise degrades both the spatial and contrast resolution of the image. So, speckle suppression or despeckling is necessary before processing like image segmentation, edge detection, and in general any medical diagnosis. It is quite a mind-numbing task to analyze the corrupted images. Among many methods that have been proposed to perform this task either in spatial domain or in transformed domain, there exists a class of approaches that use coefficient modelling in transform domain. In this paper, we proposed a novel despeckling method in the nonsubsampled shearlet transform (NSST) domain with coefficient modelling. We used Bayesian maximum a posteriori (MAP) estimator with the priori assumption as heavy-tailed Lévy (HTL) distribution for estimating the noise-free NSST coefficients. Finally, experiments show that the proposed method outperforms others in terms of visual evaluation and image assessment parameters. Manuscript profile

Iris recognition is one of the most reliable methods for identification. In general, itconsists of image acquisition, iris segmentation, feature extraction and matching. Among them, iris segmentation has an important role on the performance of any iris recognition system. Eyes nonlinear movement, occlusion, and specular reflection are main challenges for any iris segmentation method. In this paper, we propose a new pupil localization method based on the sparse representation and sparse recovery (SR). The main advantage of our segmentation algorithm based on sparse representation in respect to other approaches is capability of searching the whole image for iris region very fast. Also we have proposed a new method for enhancing the extracted iris template when the pupil boundary is noncircular, and also a new method for creating occlusion mask based on the histogram thresholding. We have compared the SR classifier and the Hamming distance (HD) with the same size dictionary and shown that using the principal component analysis (PCA) with the SR classifier makes it very faster, whereas preserves the accuracy. The achieved results are evaluated with others in terms of the recognition accuracy and the segmentation time consuming where the CASIA V4 Lamp database used. Manuscript profile