Texture Analysis and Segmentation

In my work on bottom-up image segmentation I have worked on the use of AM-FM functions for  texture analysis and segmentation, and specifically the Dominant Component Analysis (DCA) algorithm.

The DCA algorithm summarizes texture structure in terms of a 3-d descriptor, containing information about texture scale, contrast and orientation. It constitutes a simple and efficient application of the broader AM-FM image modelling paradigm. Assuming that the image can be locally modelled in terms of an monocomponent AM-FM function, the DCA method passes the image  through a multiband filterbank, demodulates each channel separately and uses the amplitude and frequency estimates of each channel as a texture descriptor.

 My research has focused on the following aspects of the problem:

• Casting the DCA  algorithm in a detection-theoretic framework, using generative models to phrase the channel selection problem.
• Building equivalent models for edges, allowing the discrimination between edges and texture. See e.g. reconstruction results using the two alternative models.
• Using the AM-FM features to drive the unsupervised segmentation of images.
• Combining the above two aspects of the problem by fusing the texture and non-texture cues for natural image segmentation.

Based on the above ideas, visually appealing results are obtained on natural images containing both textured and non-textured areas. In more recent work we have shown these methods are shown to yield systematically better segmentation results on the Berkeley segmentation benchmark, consisting of 100 natural images

A presentation of this work can be found here

Related Publications

I. Kokkinos, G. Evangelopoulos and P. Maragos,

IEEE Trans. on Pattern Analysis and Machine Intelligence (PAMI), accepted for publication

I. Kokkinos and P. Maragos,

A Detection-Theoretic Approach to Texture and Edge Discrimination.,

G. Evangelopoulos, I. Kokkinos and P. Maragos,

Advances in Variational Image Segmentation using AM-FM models: Regularized Demodulation and Probabilistic Cue Integration.,

Proc. 3rd IEEE Variational and Level-Set Methods (VLSM) Workshop, in conjunction with ICCV 05.

Advances in Texture Analysis: Energy Dominant Component & Multiple Hypothesis Testing.,

Proc. IEEE Int'l. Conf. on Image Processing, 2004.

I. Kokkinos, G. Evangelopoulos and P. Maragos,

Modulation-Feature based Textured Image Segmenation Using Curve Evolution.,

Proc. IEEE Int'l. Conf. on Image Processing, 2004.

Back to main page