SAR Change Detection Based On Double Noise Similarity Model

The redundancy is a primary character in image processing. The utilization of the redun-dancy is the key problem in many image processing fields, such as image compression, inpainting, classification, detection and so on. In order to utilize the image redundancy ef-fectively, enough similar pixels/patches are needed. Therefore, how to compare the noisy pixels/patches is a very crucial and meaningful problem. To decrease the bias in patch sim-ilarity measure, the noise similarity and its applications on SAR/PolSAR despeckling and change detection are mainly studied in this thesis. The main works are summarized as fol-lows:(1) A new concept of noise similarity (NS) is proposed to refine the comparison of noisy patch and enhance the denoising power of the nonlocal means (NLM) filter. The fact behind this concept is that the similarity of noisy patch should depend on not only the underlying signal (noise free patches), but also the noise. Based on the concept of noise similarity, we derived a double NS (DNS) model, which converts the denoising problem into the problem of reducing two kinds of noise:one is the superimposed additive noise; the other is the devi-ation error, defined as another kind of noise denoting the difference between similar pixels on their true intensities. The former corresponds to noise suppression, while the latter corre-sponds to the restoration of image details. To evaluate the effectiveness of the DNS model, we proposed an iterative version of the NLM filter, where the two noise similarities can work collaboratively in the framework of maximum a posterior. Finally, the experimental results demonstrate that the proposed approach can provide competitive performance when compared with other state-of-the-art NLM filters.(2) The proposed DNS model is expanded to suit with the multiplicative speckle in SAR images. Unlike the additive Gaussian noise, the multiplicative speckle noise in SAR images is very complicated. In order to explain the effectiveness of the proposed SAR-DNS model, a new NLM filter with SAR-DNS is designed for SAR images. Due to its MAP form, the SAR-DNS filter achieves good balance between the similarity of the speckle and the similarity of the underlying speckle-free signal. And then this balance can lead to excellent performance on both speckle smoothing and detail preservation. In addition, the role of the "double noise similarity" are discussed and compared with several state-of-the-art similarity measurements. Results are shown to demonstrate its competence when compared with the the same type algorithms.(3) A patch based change detection method for multitemporal SAR images (SAR-PCD) is proposed. Because of the inherent speckle interference in SAR system, a despeckling procedure is important for change detection. However, the residual speckle still exists in the filtered data and is unevenly distributed, which is not considered in the available change detectors. Besides that, the detection methods with despeckling procedure can not be adjust-ed based on the constant false alarm rate (CFAR). Specifically, the available SAR change detection methods are inefficient for the strong speckle interference cases due to the absence of the despeckling procedure. In order to overcome these drawbacks, an improved change detection framework for SAR images is proposed. The main innovations in the new frame-work include:(1) a despeckling procedure is added for SAR change detection, which makes the algorithm robust to strong noise; (2) a new equivalent number of looks (ENL) estimator is proposed for the PolSAR/SAR data filtered by non-local means filter; (3) the proposed method which detects the changes based on the filtered data can be adjusted with different CFARs. The experiments on both synthetic and real SAR datasets have demonstrated that the effectiveness of the proposed SAR-PCD algorithm.(4) The proposed DNS model is further expanded to suit with the Polarimetric SAR da-ta. In particular, the PolSAR-DNS similarity can be separated into two components:clean image similarity and the speckle noise similarity. The former focuses its attention on speck-le suppression, while the latter concentrates on "error" reduction and thus image detail preservation. Finally, the experiments on despeckling and change detection indicate that the proposed approach can provide better results when compared with the same type of the state-of-the-art PolSAR image similarities.(5) A new synthetic aperture radar (SAR) change detection algorithm based on stacked Fisher autoencoder (SFAE) is proposed. In the framework of SFAE, the training of the network includes unsupervised layer-wise feature learning and supervised fine-tuning. The trained network can be used to detect the changes in both of the single and multi-polarization SAR datasets in real-time without supervision. There are two innovations in this paper. The first one is to expand the stacked autoencoder to suit the environment with the multiplicative noise in SAR change detection. By introducing the Fisher discrimination, the other inno-vation is to make sure that the features extracted by the SFAE network become easier to be classified. The experiments on both of the synthetic and real SAR datasets indicate that the proposed SFAE method has a significant advantage on multitemporal (SAR/PolSAR) change detection. Specifically, the proposed SFAE method is obviously superior to the real-time methods on detection accuracy and the non-real-time methods on computational complexity.