Explore Local Variation of Spatiotemporal Texture in Videosby Independent Component Analysis

1. Motivation

Independent component analysis (ICA) is a statistical and computational techniquefor revealing hidden factors that underlie an observed dataset which are synthesized by multiple variables. Each variable is supposed to be a linear combination of some unknown latent variables. With the assumption that latent variables are nongaussian and mutually independent, ICA can successfully extract these independent components from a mixing system.

In this study, we aim to extract ICs from image texture and explore the local variation of such ICs corresponding to sequential images in a video. In a simple way, for an image, we we are trying to find an expansion of the form,

Source image= s1+ s2+ ... + sk

such that for any given image in the right hand of the equation, information about one of the coefficients gives as little information as possible about the others. In other words, they represent independent components.

By studying the similarity and difference among such ICs and Si in similar images and different images, we hope to explore that:

1. Weather ICA can find some principle characters in an image texture;

2. Weather ICA is another efficient way to compare the similarity among images;

3. Weather ICA can be successfully used to detect motion image in a video;

4. Weather ICA can be successfully used to detect motion direction of a object in a video.

2. Proposed Methods

Experiment1: Explore the ICs of an image texture

Step1: An image is represented by a vector with 1728 elements

Step2: With 10 neighborhood images, we have a image matrix

Step3: Data normalization

Step4: By FastICA algorithm, we detect the most important ICs in the matrix

Step5: Compare the ICs and principle components acquired by PCA, we explore that if there is some underlying factors, which can be detected by ICA, but failed by PCA.

Experiment2: Detect the motions in a video

Step1: Compare the ICs between similar images and different images

Step2: Select a similarity measure

Step3: Detect the motion in a video by comparing ICs

Experiment3: Detect the motion direction in a video

Step1: Divide each image into several blocks;

Step2:

Step2: For each block, we find ICs with the same way in Experiment1

Step3: When a object moves in sequential images, we expect to see the move of some IC in corresponding ICs

Step4: By setting up a motion measure, we expect to detect the object moving by ICs

Step5: By the location of blocks, we can measure the motion direction

Note: This is our first practice to apply ICA on image texture and can not grantee all of our study objectives are smoothly gained. During the study procedure, if we meet some difficulties, we will contact Dr. Latecki and ask for his guidance.

3. Datasets

We will use the Activity Matrix at

4. References

1) A. Hyvärinen and E. Oja. Independent Component Analysis: Algorithms and Applications. Neural Networks, 13(4-5):411-430, 2000.

2) D. Pokrajac and L. J. Latecki: Spatiotemporal Blocks-Based Moving Objects Identification and Tracking, IEEE Visual Surveillance and Performance Evaluation of Tracking and Surveillance (VS-PETS), October 2003.

3) L. J. Latecki and D. Pokrajac. Entropy-Based Approach for Detecting Feature Reliability. Invited Paper, 48th Annual Conf. FOR ELECTRONICS, TELECOMMUNICATIONS, COMPUTERS, AUTOMATION, AND NUCLEAR ENGINEERING (ETRAN). Cacak, Serbia, 2004.

4) P.O. Hoyer and A. Hyvärinen. Independent Component Analysis Applied to Feature Extraction from Colour and Stereo Images. Network: Computation in Neural Systems, 11(3):191-210, 2000.

5) Longin Jan Latecki, Roland Miezianko, Dragoljub Pokrajac. Motion Detection Based on Local Variation of Spatiotemporal Texture

6) Longin Jan Latecki, Dragoljub Pokrajac , Roland Miezianko. Instantaneous Reliability Assessment of Motion Features inSurveillance Videos