Research On Performance Assessment For Despeckling Algorithm Of SAR Image

Speckle reduction is a critical preprocessing step in SAR image processing. It is inevitable that the existing algorithms for despeckling have been weakened the structural features. However, the loss of structural information has seriously affected the effective extraction of structural features. Consequently, it is important to evaluate the despeckling algorithms from the structure level. Subjective assessments are not expanded since they were expensive and time-consuming. The traditional objective metrics are easy to copy but fail to keep a consistency with visual assessment. On the basis of deeply studying the traditional objective evaluation methods, a new evaluation metric for evaluating the performance of despeckling algorithm is proposed in this thesis. The main contents are as follows:(1) A full-reference performance assessment called despeckling structure loss (DSL) is proposed for evaluating the despeckling algorithm based on edge detection. The ratio image often contains residual image information, which is the difference between the SAR images and its despeckled results. According to the residual image in ratio image, by taking into account characteristics of the best and the worst structure preserving in despeckling, the proposed metric examines the presence of structural features in ratio images by using local correlations between the ratio image and the noise-free reference image at edge points.(2) Based on the above proposed metric, a no-reference performance assessment for evaluating the despeckling algorithm based on ratio image is presented. By taking into account residual information in ratio image and the SAR image, the new metric can examine the loss of SAR image by using the similarity between the SAR images and the ratio image at each pixel. The no-reference metric can evaluate and compare the speckle reduction methods and has a promising application.The despeckling algorithms are applied to synthetic SAR images and real SAR images and ratio images are obtained, and finally image quality is evaluated by using the despeckling performance evaluation methods proposed in this thesis. The proposed metrics have been tested on despeckled results of simulated as well as real SAR images and the efficiency has been demonstrated compared with the commonly used despeckling metrics. The proposed metrics can compare the performance of SAR image despeckling algorithm and keep a consistency with the structure loss shown in despeckled results as well as ratio images.