Hongxiao Feng

SAR Image Despeckling Based on Local Homogeneous-Region Segmentation by Using Pixel-Relativity Measurement

This paper provides a novel pointwise-adaptive speckle filter based on local homogeneous-region segmentation with pixel-relativity measurement. A ratio distance is proposed to measure the distance between two speckled-image patches. The theoretical proofs indicate that the ratio distance is valid for multiplicative speckle, while the traditional Euclidean distance failed in this case. The probability density function of the ratio distance is deduced to map the distance into a relativity value. This new relativity-measurement method is free of parameter setting and more functional compared with the Gaussian kernel-projection-based ones. The new measurement method is successfully applied to segment a local shape-adaptive homogeneous region for each pixel, and a simplified strategy for the segmentation implementation is given in this paper. After segmentation, the maximum likelihood rule is introduced to estimate the true signal within every homogeneous region. A novel evaluation metric of edge-preservation degree based on ratio of average is also provided for more precise quantitative assessment. The visual and numerical experimental results show that the proposed filter outperforms the existing state-of-the-art despeckling filters.

SAR Image Despeckling Based on Nonsubsampled Shearlet Transform

Synthetic Aperture Radar (SAR) Image despeckling is an important problem in SAR image processing since speckle may interfere with automatic interpretation. This paper presents a new approach for despeckling based on nonsubsampled shearlet transform. The approach introduced here presents two major contributions: (a) Translation-invariant Nonsubsampled Shearlet Transform (NSST) is designed to get more directional subbands which help to capture the anisotropic information of SAR image, and an estimation of speckle variance based on NSST is modeled to shrink NSST coefficients; (b) NSST coefficients are divided into several classes based on NSST- Multiscale Local Coefficient Variation (NSST-MLCV), which is helpful to reduce the undesired over-shrinkage, and shrinkage factor is obtained by computing the prior ratio and the likelihood ratio through mask. This model allows us to classify the NSST coefficients into classes having different degrees of heterogeneity, which can reduce the shrinkage ratio for heterogeneity regions while suppresses speckle effectively to realize both despeckling and detail preservation. Experimental results, carried out on both artificially speckled images and true SAR images, demonstrate that the proposed filtering approach outperforms the previous filters, irrespective of the features of the underlying reflectivity.

Hybrid multiobjective estimation of distribution algorithm by local linear embedding and an immune inspired algorithm

A novel hybrid multiobjective estimation of distribution algorithm is proposed in this study. It combines an estimation of distribution algorithm based on local linear embedding and an immune inspired algorithm. Pareto set to the continuous multiobjective optimization problems, in the decision space, is a piecewise continuous (m-1)-dimensional manifold, where m is the number of objectives. By this regularity, a local linear embedding based manifold algorithm is introduced to build the distribution model of promising solutions. Besides, for enhancing local search ability of the EDA, an immune inspired sparse individual clone algorithm (SICA) is introduced and combined with the EDA. The novel hybrid multiobjective algorithm, named HMEDA, is proposed accordingly. Compared with three other state-of-the-art multiobjective algorithms, this hybrid algorithm achieves comparable results in terms of convergence and diversity. Besides, the tradeoff proportions of EDA to SICA in HMEDA are studied. Finally, the scalability to the number of decision variables of HMEDA is investigated too.

SAR images reconstruction based on Compressive Sensing

Chirp signals are transmitted by Synthetic Aperture Radar (SAR) and the received signals are sampled into Inphase and Quadrature components which are so-called raw SAR data. The data is so tremendous that it brings extraordinarily high burden to the on-board storage and downlink bandwidth. This paper addresses a new process of the raw SAR data by sampling the data below Nyquist rate in terms of Compressive Sensing, which shows that super-resolved data can be reconstructed from an extremely small set of measurements than what is generally considered necessary. A wavelet-based contourlet transform, a multi-scale random Gaussian sampling, and a stage-wise directional pursuit are cooperating in this new process framework to realize our purpose, and it turns out that with only above 20% of the original transmission data should we reconstruct SAR images promisingly. Two major improvements of this radar transmission system are achieved: (a) potentially low “information rate” is preferred rather than high Nyquist rate while the transmission end emit the raw data, and (b) the hardware is significantly alleviated that lots of resources and energy can be saved in the manufacture process. This idea could enable the alleviation of transmission burden, reducing the sampling rates, the transmission time, the measurement time dramatically, shifting the emphasis from expensive transmission hardware to three smart gradients of CS framework.

Correction to “SAR Image Despeckling Based on Nonsubsampled Shearlet Transform” [Jun 12 809-823]

We regret that the PSNRs of our method in Table III in the above titled paper (ibid., vol. 5, no. 3, pp. 809-823, Jun. 2012), were left out. The correct Table III is presented here.

Image denoising using improved adaptive proportion-shrinking algorithm based on second generation bandelets

As one important multiresoltion geometry analysis tool, second generation bandelets can make full use of intrinsic geometry regularity of images, and then produces a sparse representation. This paper proposes a new denoising method, which is based on second generation bandelets and improved adaptive proportion-shrinking algorithm. Experiments on natural images with additive Gaussian white noise show that our method not only has the high peak signal to noise ratio(PSNR) value, but also has finer impression in vision, especially, has better performance on preservation of edges information and textures information than the classical proportion-shriking algorithm.