Hiep Luong

A fast non-local image denoising algorithm

In this paper we propose several improvements to the original non-local means algorithm introduced by Buades et al. which obtains state-of-the-art denoising results. The strength of this algorithm is to exploit the repetitive character of the image in order to denoise the image unlike conventional denoising algorithms, which typically operate in a local neighbourhood. Due to the enormous amount of weight computations, the original algorithm has a high computational cost. An improvement of image quality towards the original algorithm is to ignore the contributions from dissimilar windows. Even though their weights are very small at first sight, the new estimated pixel value can be severely biased due to the many small contributions. This bad influence of dissimilar windows can be eliminated by setting their corresponding weights to zero. Using the preclassification based on the first three statistical moments, only contributions from similar neighborhoods are computed. To decide whether a window is similar or dissimilar, we will derive thresholds for images corrupted with additive white Gaussian noise. Our accelerated approach is further optimized by taking advantage of the symmetry in the weights, which roughly halves the computation time, and by using a lookup table to speed up the weight computations. Compared to the original algorithm, our proposed method produces images with increased PSNR and better visual performance in less computation time. Our proposed method even outperforms state-of-the-art wavelet denoising techniques in both visual quality and PSNR values for images containing a lot of repetitive structures such as textures: the denoised images are much sharper and contain less artifacts. The proposed optimizations can also be applied in other image processing tasks which employ the concept of repetitive structures such as intra-frame super-resolution or detection of digital image forgery.

Augmented Lagrangian based reconstruction of non-uniformly sub-Nyquist sampled MRI data☆

Abstract MRI has recently been identified as a promising application for compressed-sensing-like regularization because of its potential to speed up the acquisition while maintaining the image quality. Thereby non-uniform k-space trajectories, such as random or spiral trajectories, are becoming more and more important, because they are well suited to be used within the compressed-sensing (CS) acquisition framework. In this paper, we propose a new reconstruction technique for non-uniformly sub-Nyquist sampled k-space data. Several parts make up this technique, such as the non-uniform Fourier transform (NUFT), the discrete shearlet transform and a augmented Lagrangian based optimization algorithm. Because MRI images are real-valued, we introduce a new imaginary value suppressing prior, which attenuates imaginary components of MRI images during reconstruction, resulting in a better overall image quality. Further, a preconditioning based on the Voronoi cell size of each NUFT data point speeds up the conjugate gradient optimization used as part of the optimization algorithm. The resulting algorithm converges in a relatively small number of iterations and guarantees solutions that fully comply to the imposed constraints. The results show that the algorithm is applicable not only to sub-Nyquist sampled k-space reconstruction, but also to MR image fusion and/or resolution enhancement.

Sparse representation and position prior based face hallucination upon classified over-complete dictionaries

In compressed sensing theory, decomposing a signal based upon redundant dictionaries is of considerable interest for data representation in signal processing. The signal is approximated by an over-complete dictionary instead of an orthonormal basis for adaptive sparse image decompositions. Existing sparsity-based super-resolution methods commonly train all atoms to construct only a single dictionary for super-resolution. However, this approach results in low precision of reconstruction. Furthermore, the process of generating such dictionary usually involves a huge computational cost. This paper proposes a sparse representation and position prior based face hallucination method for single face image super-resolution. The high- and low-resolution atoms for the first time are classified to form local dictionaries according to the different regions of human face, instead of generating a single global dictionary. Different local dictionaries are used to hallucinate the corresponding regions of face. The patches of the low-resolution face inputs are approximated respectively by a sparse linear combination of the atoms in the corresponding over-complete dictionaries. The sparse coefficients are then obtained to generate high-resolution data under the constraint of the position prior of face. Experimental results illustrate that the proposed method can hallucinate face images of higher quality with a lower computational cost compared to other existing methods.

A GPU-Accelerated Real-Time NLMeans Algorithm for Denoising Color Video Sequences

The NLMeans filter, originally proposed by Buades et al., is a very popular filter for the removal of white Gaussian noise, due to its simplicity and excellent performance. The strength of this filter lies in exploiting the repetitive character of structures in images. However, to fully take advantage of the repetitivity a computationally extensive search for similar candidate blocks is indispensable. In previous work, we presented a number of algorithmic acceleration techniques for the NLMeans filter for still grayscale images. In this paper, we go one step further and incorporate both temporal information and color information into the NLMeans algorithm, in order to restore video sequences. Starting from our algorithmic acceleration techniques, we investigate how the NLMeans algorithm can be easily mapped onto recent parallel computing architectures. In particular, we consider the graphical processing unit (GPU), which is available on most recent computers. Our developments lead to a high-quality denoising filter that can process DVD-resolution video sequences in real-time on a mid-range GPU.

Efficient design of a low redundant Discrete Shearlet Transform

Recently, there has been a huge interest in multiresolution representations that also perform a multidirectional analysis. The Shearlet transform provides both a multiresolution analysis (such as the wavelet transform), and at the same time an optimally sparse image-independent representation for images containing edges. Existing discrete implementations of the Shearlet transform havemainly focused on specific applications, such as edge detection or denoising, and were not designed with a low redundancy in mind (the redundancy factor is typically larger than the number of orientation subbands in the finest scale). In this paper, we present a novel design of a Discrete Shearlet Transform, that can have a redundancy factor of 2.6, independent of the number of orientation subbands, and that has many interesting properties, such as shift-invariance and self-invertability. This transform can be used in a wide range of applications. Experiments are provided to show the improved characteristics of the transform.

Robust reconstruction of low-resolution document images by exploiting repetitive character behaviour

In this paper, we present a new approach for reconstructing low-resolution document images. Unlike other conventional reconstruction methods, the unknown pixel values are not estimated based on their local surrounding neighbourhood, but on the whole image. In particular, we exploit the multiple occurrence of characters in the scanned document. In order to take advantage of this repetitive behaviour, we divide the image into character segments and match similar character segments to filter relevant information before the reconstruction. A great advantage of our proposed approach over conventional approaches is that we have more information at our disposal, which leads to a better reconstruction of the high-resolution (HR) image. Experimental results confirm the effectiveness of our proposed method, which is expressed in a better optical character recognition (OCR) accuracy and visual superiority to other traditional interpolation and restoration methods.

Image interpolation using constrained adaptive contrast enhancement techniques

In this paper we present a method for interpolating images that also preserves sharp edge information. We concentrate on tackling blurred edges by mapping level curves of the image. Level curves or isophotes are spatial curves with constant intensity. The mapping of these intensities can be seen as a local contrast enhancement problem, therefore we can use contrast enhancement techniques coupled with additional constraints for the interpolation problem. A great advantage of this approach is that the shape of the level set contours is preserved and no explicit edge detection is needed here. Results show an improvement in visual quality: edges are sharper and ringing effects are removed.

Joint photometric and geometric image registration in the total least square sense

This paper presents a novel robust image alignment technique that performs joint geometric and photometric registration in the total least square (TLS) sense. Therefore, we employ the total least square metric instead of the ordinary least square (OLS) metric, which is commonly used in the literature. While the OLS model is sufficient to tackle geometric registration problems, it gives no mutually consistent estimates when dealing with photometric deformations. By introducing a new TLS model, we obtain mutually consistent parameters. Experimental results show that our method is indeed more consistent and accurate in presence of noise compared to existing joint registration algorithms.

Non-Local Image Interpolation

In this paper we present a novel method for interpolating images and we introduce the concept of non-local interpolation. Unlike other conventional interpolation methods, the estimation of the unknown pixel values is not only based on its local surrounding neighbourhood, but on the whole image (non-locally). In particularly, we exploit the repetitive character of the image. A great advantage of our proposed approach is that we have more information at our disposal, which leads to better estimates of the unknown pixel values. Results show the effectiveness of non-local interpolation and its superiority at very large magnifications to other interpolation methods.

An improved HDR image synthesis algorithm

In high dynamic range (HDR) imaging, multiple photographs with different exposure times are combined into a radiance map, which reflects the radiance in real-life scenes. This involves recovering the response function of the imaging process. The technique proposed by Debevec and Malik is a well-known HDR image synthesis algorithm, but the computational complexity is relatively high, which limits the possible image size and the reconstruction quality. In this paper we present an improved joint optimization technique for estimating the camera response function (CRF) and the radiance map and a new sequential two-step optimization technique, which first estimates the CRF and then reconstructs the radiance map, resulting in better visual results and a higher SNR in remarkably less computation time.