Weighted Variance Based Scale Adaptive Threshold for Despeckling of Medical Ultrasound Images Using Curvelets

Medical ultrasound images suffer from multiplicative speckle noise that degrades the image quality, thus making automatic image analysis difficult. Many despeckling methods have been reported till date for ultrasound images, but most of them fail to enhance curved edges as these inhibit smoothing near the edges. In this paper, an adaptive technique for image denoising via thresholding of curvelet coefficients using a weighted window is proposed. The proposed scale adaptive threshold exploits the intra-band dependencies present in the curvelet coefficients using weighted variance for every pixel of the curvelet wedges except at the coarse scale. The weights of local window are proposed based on predominant directional correlations for reliable statistical dependencies in curvelet domain. In addition, a tuning parameter, image tuner is devised to assist the radiologists to control the degree of smoothness of the processed image. The proposed technique is evaluated on simulated as well as on real ultrasound images with the help of objective quality measures and clinical evaluations, respectively. Experimental results show that the proposed technique exhibits substantial improvement in terms of all quality metrics when compared with spatial filters, wavelet and curvelet based counterparts. Evaluations show that the proposed technique suppresses speckle effectively while preserving the fine details that are essential for better diagnosis and image analysis.