Dressed Human Model DetectionHW AssignmentCAP 5937 – HWDr. Lobo / Adam SchwarzAssigned: June 22, 2004Dressed Human Model DetectionHW AssignmentPurpose:This Assignment will demonstrate an understanding of the Short-Cut Rule as described in Liang Zhao’s thesis paper and how it isapplied to de-composing contours into recognizable body parts.Assignment:You will need to implement the Short-Cut Rule and algorithm fordecomposing the contour of a person or other object. Your programshould first decide where the NCM’s (Negative Curvature Minima) arelocated and then determine the best cut for each NCM. Then as a finalstep merge the regions to create only the significant body parts.Algorithm:1. Load Contour Data (image pixels) into data structure2. Determine Location of significant NCM (tweak the thresholdvalue to match the test run data depending how you implement theNCM finding)3. Use the equations described in the paper to determine thelocations of the best cuts for EACH NCM4. (Optional ~ Bonus Points) Group the over-segmented partsleaving only the significant parts and the parts that can’tbe decomposed any further5. Display on screen the cuts along with the original contoursilhoutte6. Save this image file as output#.ext as it corresponds to theinput file: test#.rawTest Data:All test files and the example file are 128x128 b&w image filessaved in the .raw file format. They are included on the class websitein the testData.zip file, which contains the 4 test images and theexample image (example.raw) that will be used to match the examplerun (see below).Example Run:NCM Identification:First thing to check is to make sure that your NCM points arematching up closely with the above picture, it doesn’t have to beexact, but you should have the same number of points and inapproximately the same locations.All Cuts Found:Use the two equations presented in the paper for finding theoptimal cuts (see Short-Cut Rule, section 2.3 of the paper andslides 9-15). These results should look something like the middlepicture above. Note: If you choose to not do the bonusimplementation, then save this image as your output image.Final Output:Group the over-segmented parts using the method described insection 2.3.2 of the paper and slide 13 or another method of yourchoice.* This output should be saved out as an image to besubmitted with your report/summary.*Note: If you choose to use a different method for merging the over-segmented parts, describe this method in detail and add it on as anextra page to your summary report.To Turn in:You should submit your program written in the language of yourchoice along with your 4 output image files. There should also be abrief written summary talking about your experience with this project,implementation of the algorithm(s), and your results including howthey matched or didn’t match what you expected.- Program (programming language of your choice)- 4 output image files (format of your choice)- Brief summary (1-2 pages, see description above for moredetails)Test Files:test1.raw test2.rawtest3.raw
View Full Document
Unlocking...