Bayesian Denoising of Ultrasound Image Based on Wavelet Domain Hidden Markov Tree Model
基于小波域隐马尔可夫树模型的超声图象贝叶斯去噪

One of the hot key in medical image processing is how to suppressing speckle noise in ultrasound image. The low quality of ultrasound imaging brings some difficulty to sequential image processing and image analysis due to the effect of its inherent speckle noise. In this paper, a speckle suppression method for medical ultrasound image is presented. First, the logarithmic transform of the original image is analyzed into the multiscale wavelet domain. Then, the wavelet domain multiscale representation of image is regarded as Hidden Markov Tree model. The model is trained by the efficient EM algorithm,which is called Baum weltch algorithm. Speckle noise of ultrasound image is reduced by Bayesian estimator based on Hidden Markov model. Finally, the invert discrete wavelet transform and the exponent transform of the estimated wavelet coefficients obtain the denoised image in turn. Performance of the proposed method has been tested on ultrasound image. The results show the method effectively reduces the speckle while preserving the edges of the original image. Current state of the art methods, such as soft and hard thresholding, are applied on actual ultrasound medical images and compared with the novel method. The achieved performance improvement is quantified. The experiment results show the proposed method is feasible and reasonable.