Qizhi Teng

Single Image Super-Resolution Using Local Geometric Duality and Non-Local Similarity

Super-resolution (SR) from a single image plays an important role in many computer vision applications. It aims to estimate a high-resolution (HR) image from an input low- resolution (LR) image. To ensure a reliable and robust estimation of the HR image, we propose a novel single image SR method that exploits both the local geometric duality (GD) and the non-local similarity of images. The main principle is to formulate these two typically existing features of images as effective priors to constrain the super-resolved results. In consideration of this principle, the robust soft-decision interpolation method is generalized as an outstanding adaptive GD (AGD)-based local prior. To adaptively design weights for the AGD prior, a local non-smoothness detection method and a directional standard-deviation-based weights selection method are proposed. After that, the AGD prior is combined with a variational-framework-based non-local prior. Furthermore, the proposed algorithm is speeded up by a fast GD matrices construction method, which primarily relies on the selective pixel processing. The extensive experimental results verify the effectiveness of the proposed method compared with several state-of-the-art SR algorithms.

Single Image Super-Resolution via Adaptive Transform-Based Nonlocal Self-Similarity Modeling and Learning-Based Gradient Regularization

Single image super-resolution (SISR) is a challenging work, which aims to recover the missing information in an observed low-resolution (LR) image and generate the corresponding high-resolution (HR) version. As the SISR problem is severely ill-conditioned, effective prior knowledge of HR images is necessary to well pose the HR estimation. In this paper, an effective SISR method is proposed via the local structure-adaptive transform-based nonlocal self-similarity modeling and learning-based gradient regularization (LSNSGR). The LSNSGR exploits both the natural and learned priors of HR images, thus integrating the merits of conventional reconstruction-based and learning-based SISR algorithms. More specifically, on the one hand, we characterize nonlocal self-similarity prior (natural prior) in transform domain by using the designed local structure-adaptive transform; on the other hand, the gradient prior (learned prior) is learned via the jointly optimized regression model. The former prior is effective in suppressing visual artifacts, while the latter performs well in recovering sharp edges and fine structures. By incorporating the two complementary priors into the maximum a posteriori-based reconstruction framework, we optimize a hybrid L1- and L2-regularized minimization problem to achieve an estimation of the desired HR image. Extensive experimental results suggest that the proposed LSNSGR produces better HR estimations than many state-of-the-art works in terms of both perceptual and quantitative evaluations.

Single image super resolution using local smoothness and nonlocal self-similarity priors

Abstract Single image super resolution (SISR) is an inverse problem, so an effective image prior is necessary to reconstruct a high resolution (HR) image from a single low resolution (LR) image. On the one hand, natural images satisfy the property of local smoothness; on the other hand, the patches could find some similar patches in different locations within the same image, and this property is known as nonlocal self-similarity. In this paper, we propose a SISR method by incorporating the local smoothness and nonlocal self-similarity priors in the reconstruction-based SISR framework simultaneously, and the Split Bregman Iteration (SBI) optimization algorithm is imitated to solve the L1-regularized problem. Experimental results show that, in most case, the proposed method quantitatively and qualitatively outperforms the state-of-the-art SISR algorithms.

Rotation expanded dictionary-based single image super-resolution

In this report, issues that affect the performance of the neighbor embedding (NE)-based Super-Resolution (SR) method are analyzed. Effective enrichment of the dictionary is a critical factor for the NE-based SR method. To efficiently enhance the dictionarys expressive capability, a rotation expanded dictionary (RED) incorporating the Radon transform (RT) technique is proposed. By representing patch rotations with a compact scheme, both the search for neighbors and the estimation of rotation angles in the SR process are significantly simplified. To refine the patch matching accuracy when using the expanded dictionary, a new level of imaging, known as the middle-resolution (MR) image, is proposed to replace the original low-resolution (LR) image in patch matching. Because MR patches bear more distinguishable features, this modification is able to identify neighbors more accurately for the input patches. Lastly, the effects of a single image SR method based on the MR matching and the rotation expansion are examined in simulations. A comprehensive comparison with several state-of-the-art SR methods demonstrates the superior performance of the proposed method. Expanding the dictionary by rotation deformation effectively enriches the dictionary abundance.Radon transform largely simplifies the use of rotation expanded dictionary (RED).A so-called mid-resolution (MR) image is constructed to address the ambiguity problem and improve the matching accuracy.The proposed SR incorporating both RED and MR matching produces superior performance.

Low bit rates image compression via adaptive block downsampling and super resolution

Abstract. A low bit rates image compression framework based on adaptive block downsampling and super resolution (SR) was presented. At the encoder side, the downsampling mode and quantization mode of each 16×16 macroblock are determined adaptively using the ratio distortion optimization method, then the downsampled macroblocks are compressed by the standard JPEG. At the decoder side, the sparse representation-based SR algorithm is applied to recover full resolution macroblocks from decoded blocks. The experimental results show that the proposed framework outperforms the standard JPEG and the state-of-the-art downsampling-based compression methods in terms of both subjective and objective comparisons. Specifically, the peak signal-to-noise ratio gain of the proposed framework over JPEG reaches up to 2 to 4 dB at low bit rates, and the critical bit rate to JPEG is raised to about 2.3 bits per pixel. Moreover, the proposed framework can be extended to other block-based compression schemes.

Space-time super-resolution with patch group cuts prior

We address space-time super-resolution (SR) problem in this paper. To efficiently explore the correlations within video sequences, a patch group (PG) model is proposed. The model is based on a novel 3D neighborhood system (NS) and embeds the spatial and temporal correlations. A patch group cuts (PGCuts) metric is then built on the model to provide a new prior for space-time SR reconstruction. To balance the prior strength for the whole video sequence, an adaptive scheme is also considered. We evaluate the proposed PGCuts prior on both synthesized and real video sequences. The results indicate that space-time SR method with the PGCuts prior outperforms other ones by retaining smooth edges and motion trajectories in videos, and by staying robust to noises as well. 3D neighborhood system accommodates more pixel correlations than traditional one.Patch group model based on similarity efficiently exploits space-time coherence.Smooth surfaces in patch group model results in smooth edges and motion trajectories.Edge strength based adaptive scheme effectively balances performance in the whole video.SR with patch group cuts produces better objective and subjective performance.

Color image segmentation algorithm based on neural networks

This paper presents a color image segmentation method with Self-Organize Feature Map and General Learning Vector Quantity which, in the uniform color space, divides color into clusters based on the least sum of squares criterion. At the first step of this method, SOFM is employed to make a preliminary classification on the original image, and then GLVQ is used to segment it. Both of their advantages can be fully taken of to improve the precision and velocity of color image segmentation.

Autofocus methods for automated microscopy

This paper introduced a basis principle of an autofocus system, including the hardware composition and the software flow chart. The most important work of an autofocus system is to construct or select a suitable focus function. We first discussed some traditional focus functions, pointed out their advantages and shortages. Then we proposed a new autofocus algorithm, which based on wavelet transform. We use the diagonal edge information of the result of the wavelet transform to form focus function. After many experiments we draw a conclusion that the wavelet transform method is superiority than others, use it can get a highly accuracy along with a fast speed, special in the biomedical microscopic image analysis. At the last of this paper we discussed the further work related to the improvement of hardware condition and the optimization of focus function.

SGCRSR: Sequential gradient constrained regression for single image super-resolution

Abstract Single image super-resolution (SISR), which aims to produce an image with higher resolution and better visual quality from the given single low-resolution (LR) image, has attracted extensive attention in recent years. In particular, the regression-based SISR approaches, which learn the mapping between LR and high-resolution (HR) patch pairs, are efficient and effective as a whole. However, the super-resolved images produced by this kind of method often suffer from visual artifacts as no extra constraints or priors are enforced. To alleviate these shortcomings, we propose a Sequential Gradient Constrained Regression-based single image Super-Resolution (SGCRSR) framework, which provides an effective way to combine the conventional learning-based and reconstruction-based approaches. Firstly, we improve the performance of the well-known super-resolution (SR) method A+ by addressing its deficiencies in both training and testing stages and propose the enhanced A+ (EA+). Then, the EA+ model is applied in dual intensity–gradient domain to construct the Gradient Constrained Regression (GCR)-based SR unit. Finally, a GCR-based sequential SR framework, namely the SGCRSR, is established to improve the quality of super-resolved images gradually. Extensive experiments show that the proposed SGCRSR achieves better performance than several state-of-the-art SR algorithms.

Similarity Evaluation of 3D Gray Rock Image Using Pattern Density Classification Function

Aiming at the problem that the existing 3D core similarity evaluation methods cannot effectively evaluate gray core images, we proposed a similarity evaluation algorithm based on Pattern Density Classification Function (PDCF). First of all, the 3D template is used to extract the texture patterns of 3D core images, and then the pattern density classification function is formed with the extracted patterns by adopting K-means algorithm. An adaptive method is used to find out the appropriate K value. Finally, the pattern density classification function is used to measure the texture similarity between 3D gray rock models. In this paper, a comparative experiment of multiple groups of core images is carried out. Combined with the existing similarity evaluation algorithm for the binary image of 3D core, the 3D gray core model similarity characterization is realized from morphological and texture distribution. Keywords—3D rock models; pattern density classification function; morphological similarity; texture similarity