Adib Akl

Texture Synthesis Using the Structure Tensor

This paper proposes a two-stage texture synthesis algorithm. At the first stage, a structure tensor map carrying information about the local orientation is synthesized from the exemplars data and used at the second stage to constrain the synthesis of the texture. Keeping in mind that the algorithm should be able to reproduce as faithfully as possible the visual aspect, statistics, and morphology of the input sample, the method is tested on various textures and compared objectively with existing methods, highlighting its strength in successfully synthesizing the output texture in many situations where traditional algorithms fail to reproduce the exemplars patterns. The promising results pave the way towards the synthesis of accurately large and multi-scale patterns as it is the case for carbon material samples showing laminar structures, for example.

Structure tensor based synthesis of directional textures for virtual material design

Exemplar-based texture synthesis schemes are promising for virtual material design. They provide impressive results in many cases, but fail in difficult situations with large and multi-scale patterns, or with long range directional variations. Since a prior synthesis of a geometric layer may help in the synthesis of the texture layer, a two-stage structure/texture synthesis algorithm is proposed. At the first stage, a structure tensor map carrying information about the local orientation is synthesized from the exemplars data, and at the second stage, the synthesized tensor field is used to constrain the synthesis of the texture. Results show that the proposed approach not only yields better synthesized textures, but also successfully synthesizes the output texture in many situations where traditional algorithms fail to reproduce the exemplars patterns, which paves the way towards the synthesis of accurately large and multi-scale patterns as it is the case for pyrolytic carbon samples showing laminar structures observed by Transmission Electronic Microscopy.

Structure tensor regularization for texture analysis

Texture analysis has been a particularly dynamic field with different computer vision and image processing applications. Most of the existing texture analysis techniques yield to significant results in different applications but fail in difficult situations with high sensitivity to noise. Inspired by previous works on texture analysis by structure layer modeling, this paper deals with representing the textures structure layer using the structure tensor field. Based on texture pattern size approximation, the proposed algorithm investigates the adaptability of the structure tensor to the local geometry of textures by automatically estimating the sub-optimal structure tensor size. An extension of the algorithm targeting non-structured textures is also proposed. Results show that using the proposed tensor size regularization method, relevant local information can be extracted by eliminating the need of repetitive tensor field computation with different tensor size to reach an acceptable performance.

Edge-Based Automatic Parameter Selection for Sub-Optimal Image Despeckling Using Kuan Filter

Multiplicative speckle noise is a common problem found in several imaging applications, mainly in SAR and ultrasound imaging. This paper targets the Kuan despeckling filter and seeks to optimize its output without the need for several runs of the filter with the parameters adjusted each time. The proposed solution estimates the optimal filter parameter value, which results in near-optimal performance, where the PSNR loss does not exceed 0.1 dB most of the time, compared to the best possible filter output.

Synthesis of arbitrary-shaped textures constrained by the structure tensor field

In this paper, a new algorithm is presented for the nonparametric synthesis of arbitrary-shaped textures from an initial texture sample, called the exemplar, and a reference orientation map, called the reference and intended to constrain the coarse scale orientation flow of the output texture. The synthesis process consists of three stages. First, a dictionary is constructed, composed of rotated versions of the exemplar and of their structure tensor fields. Then a synthetic structure tensor field is constructed from the exemplar’s one, under the constraint of the reference. Finally, the output texture is synthesized from the previously generated structure tensor field, again under the constraint of the reference. Synthesis experiments show that the proposed method is able to faithfully reproduce the visual aspect of input samples while respecting the large-scale orientation flow of the reference. This paves the way toward the synthesis of 3-D textures of arbitrary shapes from 2-D exemplars, with applications in virtual material design.

Second-moment matrix adaptation for local orientation estimation

This paper addresses the orientation estimation in digital images using gradient-based methods. In particular, the second-moment matrix is used to extract the information about the local orientation and the degree of anisotropy in the image, mainly in structured and sinusoid-like textured images. Keeping in mind that the extent of gradient fields smoothing should be decent to extract as faithfully as possible the local orientation information, an algorithm for the estimation of the Gaussian kernel size used in computing the second-moment matrix is proposed. The results obtained on various textured images highlight the strength of the proposed approach in successfully extracting the local variation of orientations in the underlying image, which paves the way towards the accurate extraction of image structures, used in different applications like image synthesis, for example.

A hybrid wavelet-spatial denoising filter

Noise is one of the most widespread problems present in nearly all imaging applications. The search for efficient image denoising methods is still a valid challenge. In spite of the sophistication of the recently proposed methods, most algorithms have not yet attained a desirable level of applicability. They all show an outstanding performance when the image model corresponds to the algorithm assumptions, but fail in general and create artifacts or remove fine image structures. Therefore, a universal “best” filter has yet to be found. Wavelet analysis is a new method consisting of a set basis functions that can be used to analyze signals in both time (or space) and frequency domains simultaneously. In this paper, a novel hybrid filter for image despeckling that combines wavelet denoising and an enhanced adaptive Kuan filter is proposed, resulting in a significant gain with respect to many spatial as well as wavelet-based speckle reduction filters.

A survey of exemplar-based texture synthesis methods

Abstract Texture synthesis has become an important topic in image processing, with many fundamental applications in computer vision and image understanding. The purpose of this survey is to give an overview and classification of texture synthesis approaches. According to how they represent, analyze and synthesize textures, the methods are divided into three families; procedural synthesis, exemplar-based synthesis and model-based synthesis, while focusing on exemplar-based methods including most of the synthesis techniques. Finally, experimental evaluations on different textures show that non-parametric synthesis methods lead to the best results when dealing with regular structured textures while parametric algorithms are better for the synthesis of more irregular textures.

A Tensor-based Technique for Structure-aware Image Inpainting.

Image inpainting is an active area of study in computer graphics, computer vision and image processing. Different image inpainting algorithms have been recently proposed. Most of them have shown their efficiency with different image types. However, failure cases still exist, especially when dealing with local image variations. This paper presents an image inpainting approach based on structure layer modeling, where this latter is represented by the second-moment matrix, also known as the structure tensor. The structure layer of the image is first inpainted using the non-parametric synthesis algorithm of Wei and Levoy, then the inpainted field of second-moment matrices is used to constrain the inpainting of the image itself. Results show that using the structural information, relevant local patterns can be better inpainted comparing to the standard intensity-based approach.

Structure-based image inpainting

Image inpainting is a dynamic field with different image processing and computer graphics applications. Most of the existing image inpainting methods lead to significant results in different applications but fail in difficult situations with high local structural variations. In this paper, a structure-based image inpainting algorithm is proposed, where the images structure layer is represented and analyzed using the structure tensor field. The structure layer of the image is first inpainted by adapting the Efros and Leung algorithm to the specificities of the structure tensor, then the obtained tensor field is used to help the image inpainting process. Results show that using the proposed method, relevant local information can be better inpainted comparing to the initial intensity-based approach that does not consider structural information during the inpainting process.