Texture is a one of the most fundamental yet important characteristic of images, and texture analysis & modeling is an essential and challenging problem in computer vision and pattern recognition, which has attracted extensive research attention over the last several decades.

As a powerful visual cue, texture play an important role in human perception, and provides useful information in identifying objects or regions in images, ranging from multispectral satellite data to microscopic images of tissue samples. Besides, understanding texture is also a key component in many other computer vision topics, including image de-noising, image super-resolution and image generation.

In the past few years, MCL has done original research works in several important aspects of texture analysis & modeling, including texture representation, unsupervised texture segmentation, and dynamic texture synthesis.

Texture Representation[1]: A hierarchical spatial-spectral correlation (HSSC) method is proposed for texture analysis in this work. The HSSC method first applies a multi-stage spatial-spectral transform to input texture patches, which is known as the Saak transform. Then, it conducts a correlation analysis on Saak transform coefficients to obtain texture features of high discriminant power. During the correlation analysis, both auto-correlation and cross-correlation are computed, and further used to get compact and representative feature for texture. Given the fact that texture is the spatial organization of a set of basic patterns, we further provide theoretical explanation of proposed method, that it attempts to capture the energy distribution of orthogonal texture patterns derived from the Saak transform. This paper has been accepted by ICIP2019.

Unsupervised Texture Segmentation[2]: We propose a data-centric approach to efficiently extract and represent textural information for unsupervised texture segmentation problem. Based on the strong self-similarities and quasi-periodicity in texture images, the proposed method first constructs a representative texture pattern set for the given image by leveraging the patch clustering strategy. Then, pixel-wise texture features are designed according to the similarities between local patches and the representative textural patterns. Moreover, the proposed feature is generic and flexible, and can perform segmentation task by integrating it into various segmentation approaches easily. Extensive experimental results on both textural and natural image segmentation show superiority of the proposed method. This paper has been accepted by ICASSP2019.

Dynamic Texture Synthesis: Dynamic texture synthesis aims to synthesize dynamic texture video given a short video of target dynamic texture as reference, like flame, smoke, and wavy water. Understanding and characterizing these temporal patterns has been a problem of interest in human perception, computer vision, and pattern recognition for the past several years. Comparing with the static texture image synthesis, the major challenge of this problem is about how to maintain both spatial consistency and temporal consistency in the generated samples. With the help of proposed techniques, we could generalize the baseline model from random dynamic texture synthesis to structural dynamic texture synthesis, which could substantially enlarge the feasible scope of previous method. Extensive experiments show that our model could offer the state-of-the-art synthesized results.

By Kaitai Zhang, Hong-Shuo Chen, Ye Wang, Xinfeng Zhang,  Xiangyang Ji, C.-C. Jay Kuo


[1] Zhang, Kaitai, et al. “Texture Analysis via Hierarchical Spatial-spectral Correlation (HSSC).” ICIP 2019-2019 IEEE International Conference on Image Processing (ICIP). IEEE, 2019

[2] Zhang, Kaitai, et al. “A Data-centric Approach to Unsupervised Texture Segmentation Using Principle Representative Patterns.” ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019.