Rotation, Scaling and Illumination Invariant Pattern Recognition Using Joint Transform Correlation for Object Detection and Tracking

Authors

Presenter(s)

Sidike Paheding

Files

Description

Challenges in pattern recognition mainly includes object rotation, scaling, and illumination variations. Joint Transform Correlation (JTC) based filtering techniques yield promising outputs in optical pattern recognition and they have been widely used for real-time pattern recognition applications such as object detection and tracking. However, objects in complex background brings difficulty to JTC based algorithms since the performance of the JTC is sensitive to object distortions such as changes due to rotation, scaling, and illumination. One of the solutions is to add or modify filters during JTC process. Synthetic discriminant function (SDF) can be integrated with fringe-adjusted filter to alleviate the problems of scale and rotation variations of the target. Fringe-adjusted JTC with monogenic signal representation can achieve illumination invariant pattern recognition. In the case of multiple target detection, the input-scene subtraction algorithm can be employed in JTC to efficiently detect multiple targets simultaneously with high correlation peak intensity with low false detection rate. While these techniques resolve specific problems of JTC, a full-fledged approach to equip the JTC with features that are robust to object rotation, scaling, and illumination variations is yet to be done. Therefore, our goal in this research is to reduce the sensitivity of the JTC to object distortions in the input image so that it can improve the detection efficiency in terms of sharper correlation peak intensity, narrow correlation width and higher pattern discriminability. In the proposed scheme, a local phase feature set is extracted prior to the JTC process, while the SDF is integrated with JTC during the correlation process. We evaluate our algorithm for face recognition and car tracking. Experimental results show that the proposed method yields better performance compared to alternate JTC based methods.

Publication Date

4-9-2015

Project Designation

Graduate Research

Primary Advisor

Vijayan K. Asari

Primary Advisor's Department

Electrical and Computer Engineering

Keywords

Stander Symposium project

Disciplines

Arts and Humanities | Business | Education | Engineering | Life Sciences | Medicine and Health Sciences | Physical Sciences and Mathematics | Social and Behavioral Sciences

Recommended Citation

"Rotation, Scaling and Illumination Invariant Pattern Recognition Using Joint Transform Correlation for Object Detection and Tracking" (2015). Stander Symposium Projects. 637.
https://ecommons.udayton.edu/stander_posters/637