Reseach And Applications On Sparsity Analysis Applied In Remote Sensing Image Processing

In the last decade, sparsity has emerged as one of the leading concepts in a wide range of signal processing applications, which include signal compression and encoding, signal encryption and transmission, signal source separation, signal denoising and reconstruction, feature extraction, to name only a few representative applications. Sparsity has also long been an attractive theoretical and practical signal property in many areas of applied mathematics, such as theoretical signal processing, statistical estimation, computational harmonic analysis, etc.The interest in sparsity has arisen owing to the new sampling theoiy, compressed sensing, which provides an alternative to the well-known Shannon sampling theory. Compressed sensing uses the prior knowledge that signals are sparse, whereas Shannon theory was designed for frequency band-limited signals. By establishing a direct link between sampling and sparsity, compressed sensing has had a huge impact in many scientific fields such as medical imaging, encoding and information theory, signal acquisition and image processing, geophysical and astronomical data analysis, etc. Compressed sensing acts today as wavelets did two decades ago, linking numerous researchers from different fields. By emphasizing so rigorously the importance of sparsity, compressed sensing has also cast light on all work related to sparse data representation, such as wavelet transform, curvelet transform, etc. Indeed, a signal is generally not sparse in direct space (i.e., pixel space), but it can be very sparse after being decomposed on a specific set of functions.This paper focuses on the sparsity which is applied in a series of application scenarios in the field of remote sensing. In the following chapters, I expand the sparsity analysis from the traditional method of signal processing (mainly focus on image denoising and inpainting) to the automatic classification of the target scenes in the remote sensing images. The second chapter introduces the intrinsic link between sparsity analysis and the concepts of morphological diversity. Based on this background knowledge, the third chapter to the fifth chapter discusses the specific applications of sparsity analysis, respectively.The third chapter mainly focuses on the sparsity analysis applied in remote sensing image denoising. This chapter introduces a remote sensing image denoising method based on generalized morphological component analysis (GMCA). This novel algorithm is the further extension of morphological component analysis algorithm to the blind source separation framework. The iterative thresholding strategy adopted by GMCA algorithm firstly works on the most significant features in the image, and then progressively incorporates smaller features to finely tune the parameters of whole model. Mathematical analysis of the computational complexity of GMCA algorithm is provided. Several comparison experiments with state-of-the-art denoising algorithms are reported. In order to make quantitative assessment of algorithms in experiments, Peak Signal to Noise Ratio (PSNR) index and Structural Similarity (SSIM) index are calculated to assess the denoising effect from the gray-level fidelity aspect and the structure-level fidelity aspect, respectively. Quantitative analysis on experiment results, which is consistent with the visual effect illustrated by denoised images, has proven that the introduced GMCA algorithm possesses a marvelous remote sensing image denoising effectiveness and ability. It is even hard to distinguish the original noiseless image from the recovered image by adopting GMCA algorithm through visual effect.The fourth chapter mainly focuses on the sparsity analysis applied in remote sensing image inpainting. This chapter introduces a remote sensing image inpainting method based on speed-up generalized morphological component analysis (SGMCA). Due to its capability to represent and separate the morphological diversities, GMCA algorithm is a state-of-the-art image inpainting method. SGMCA algorithm introduced in this chapter can accelerate the iterative process of GMCA algorithm. By adding some more assumptions to GMCA algorithm, SGMCA algorithm is proven as a much faster algorithm which can handle very large scale problems. Several experiments illustrate that SGMCA algorithm can recover the remote sensing images with different patterns of missing pixels. It is even hard to distinguish the original remote sensing image from the recovered image through visual effect. The PSNR and SSIM indices explain why the salient visual effect is obtained, and confirm the marvelous inpainting capability of SGMCA algorithm. Quantitative analysis on time consumption proves that SGMCA algorithm can greatly improve the iterative speed of GMCA algorithm, indeed.The fifth chapter mainly focuses on the sparsity analysis applied in classification of scenes in the remote sensing images. This chapter proposes a high-resolution satellite image classification method using morphological component analysis of texture and cartoon layers. The construction of the dictionary matrix used in the algorithm is based on independent component analysis. After the decomposition, we obtain the morphological coefficient vectors in both texture and cartoon layers which are termed as the sparse representation of the input high-resolution satellite image. By combining the features from two layers, the total probability of the target image classification is calculated out according to the maximum likelihood mechanism. Quantitative analysis on experiment results and comparisons with classic image classification algorithms have proven that the proposed classification method has better accuracy, efficiency, and performance than most of state-of-the-art classification methods.