Region-based Image Classification

CS 134 Project Proposal

Group Member : Qi Gu

Abstract:

The term image classification refers to labeling the input images into certain predefined categories and content-based image retrieval has emerged as an important area in computer vision and multimedia computing. The key techniques related to this field contain image representation and classification. In this project I will put my focus on a region-based image representation approach that helps improving the performance of image classification.

Goal:
The objective of my project is to introduce and implement a region-based approach that can better represent the image using low-level features and applying this method in image classification.

To illustrate the performance of this method, I will compare it with some of the existing image representation works and reason the differences of their performances.

Method:
The whole pipeline of my project can be split into 2 parts: image representation and classification. The main methods related to image representation include: Image segmentation which has been successfully used in content-based image analysis. These segmentation-based algorithms first partition the whole image into small blocks, say 4 * 4 pixels, then take into consideration of block’s color, texture, shape, etc as the basic features where the texture attibute is always computed by doing Wavelet Transformation. Then the simple procedure to gather these blocks into several regions is by using Clustering Algorithm, e.g. K-means.

Another method that shows how to convert the regions into an abstract feature vector to represent the image is by optimizing the Diverse Density value (Maron and Lozano-P′erez), which is an objective function, defined over the region feature space and regions can be viewed as points. A larger value of DD at a point indicates a higher probability that the point fits better with the regions from positive images than with those from negative.

After finishing mapping images into the new feature space, I will use Classification method to do the final step. Since classification is a well studied topic, the focus will not cover new design of the novel classification algorithm, but by choosing mature approaches, such as Logistic Regression, KNN.

The whole pipeline will be like the figure below:

Data set:
The image data used in this project is published by COREL Corporation. The whole collection includes several categories describing distinct topic of interest. Each category contains 100 images, all in JPEG format with size 384*256 or 256*384.
Here are some examples from the collection:

Reference:
I. Daubechies. Ten Lectures on Wavelets. Capital City Press, 1992.
J. A. Hartigan and M. A. Wong. Algorithm AS136: A k-means clustering algorithm. Applied
Statistics, 28:100–108, 1979.
A. Vailaya, M. A. T. Figueiredo, A. K. Jain, and H.-J. Zhang. Image classification for content-based indexing. IEEE Transactions on Image Processing, 10(1):117–130, 2001.
J. Z. Wang, J. Li, and G. Wiederhold. SIMPLIcity: Semantics-sensitive integrated matching forpicture libraries. IEEE Transactions on Pattern Analysis and Machine Intelligence, 23(9):947–963, 2001b.
Y. X. Chen, J.Z. Wang: Image Categorization by Learning and Reasoning with Regions. Journal of Machine Learning Research 5 (2004) 913–939
T. Athanasiadis, P. Mylonas, Y. Avrithis, and S. Kollias: Semantic Image Segmentation and Object Labeling. IEEE Transactions on Circuits and Systems for Video Technology vol.17, NO. 3, 2007

2: Final:
Implement image representation approach
Set up experiment scenario to do the classification test.
Analyze and reason the result.