250 likes | 272 Vues
This outline summarizes Jacob D’Avy’s tasks for the Spring semester, including writing updates, software utilities, and plans to move forward into the Summer. Research focus is on learning state-of-the-art segmentation techniques, parameter optimization, and performance evaluation. Utility development involves updating segmentation GUI and testing methods. The semester also includes reviews on parameter optimization and performance evaluation techniques. Collaboration opportunities and paper development for segmentation research are planned for the Summer. 8 Relevant
E N D
Outline • Semester tasks summary • Writing update • Software utilities • Moving forward/Summer
Outline • Semester tasks summary • Writing update • Software utilities • Moving forward/Summer
Tasks Writing • Parameter optimization review • Performance evaluation review (supervised & unsupervised) • Segmentation paper summaries Research focus • Learning state of the art of segmentation, parameter optimization, and performance evaluation. • Testing segmentation and parameter optimization methods • Potential collaborations with other IRIS students Utility development • Updating segmentation utility GUI • Platform for testing performance evaluation methods • System to run and visualize parameter search process
Outline • Semester tasks summary • Writing update • Software utilities • Moving forward/Summer
Writing http://imaging.utk.edu/research/jdavy/reports.htm
Outline • Semester tasks summary • Writing update • Parameter Optimization Review • Performance Evaluation Review • Software utilities • Moving forward/Summer
Parameter Optimization Review • A review of optimization methods that have been applied to finding parameters for segmentation. • List of methods contained in the review: * Reference list at end of presentation
Parameter Optimization Review • Heuristic search methods • - Generate parameters • - Segment the image • - Evaluate performance Image Generate parameters Segment Evaluate
Parameter Optimization Review • Heuristic search methods • - Generate parameters • - Segment the image • - Evaluate performance Image Generate parameters Segment Evaluate • Crucial evaluation feedback • Time consuming • Local minima
Outline • Semester tasks summary • Writing update • Parameter Optimization Review • Performance Evaluation Review • Software utilities • Moving forward/Summer
Unsupervised Performance Evaluation Review • A review of methods that rate the “goodness” of a segmentation without using ground truth. I have implemented F, Q, and Color saliency methods. * Reference list at end of presentation
Unsupervised Performance Evaluation Review • There are much less unsupervised methods than supervised. • Most evaluation measures try to fulfill the criteria: • Regions should be uniform and homogeneous. • Adjacent regions should be significantly different. • Boundaries should be simple. • Not an easy problem. Summary: R. Haralick, L. Shapiro, “Image Segmentation Techniques,” Computer Vision, Graphic, and Image Processing, vol. 29, pp. 100-132, 1985.
Outline • Semester tasks • Writing update • Software utilities • Segmentation Analysis GUI • Parameter Optimization platform • Moving forward/Summer
Utility development Segmentation Analysis GUI • Segment image for a range of parameters • Visualize segmentation results • Ability to test performance evaluation methods GUI Functionality Segmentation Evaluation New version is available on my website http://imaging.utk.edu/research/jdavy/webfiles/code/segUtil/segUtilv050411.zip
Outline • Semester tasks • Writing update • Software utilities • Segmentation Analysis GUI • Parameter Optimization platform • Moving forward/Summer
Utility development Parameter optimization testing platform • Parameter search using Tabu search • Visualization of search process D. Crevier, “Image Segmentation Algorithm Development Using Ground Truth Image Datasets,” Computer Vision and Image Understanding, vol. 112, no. 2, pp. 143-159, 2008.
Tabu search example • Tabu search can be used to find parameters for a segmentation method. • Tabu search uses a memory system to modify a neighborhood search window. • New parameter combinations are generated within the search window. 1. Generate parameter combinations 2. Segment Evaluation performance 3. Generate new search neighborhood F. Glover, “Tabu Search: A Tutorial,” Interfaces, vol. 20, 1990.
Tabu search example Input image: Segmentation method: Efficient graph based Evaluation method: Color saliency Parameter data: P. Felzenszwalb and D. Huttenlocher, “Efficient graph-based image segmentation,” International Journal of Computer Vision, vol. 59, no. 2, 2004 . G. Heidemann, “Region saliency as a measure for colour segmentation stability,” Image and Vision Computing, vol. 26, no. 2, pp. 211-227, 2008.
Tabu search example Input image: Highest CS score parameters : Segmented result using these parameters:
Outline • Semester tasks • Writing update • Software utilities • Moving forward/Summer
Moving Forward Summer Find collaboration opportunities with other IRIS students Develop and test parameter optimization idea Segmentation, parameter optimization, performance evaluation papers Tyndall? *Wedding in July