Image Analysis & Computer Vision projects

Possible Project Topics:

• 1) Recognition of geometric shapes, e.g. circles, squares, rectangles, triangles, etc.
• 2) Classification of different tools, e.g. screwdrivers, wrenches, hammers, etc.
• 3) Identification of different coins and bills.
• 4) Design a program to read bar codes.
• 5) Character recognition.

Other projects of your choice with emphasis on applications of pattern recognition/classification.

Important Note: Your project needs to work with real world input, i.e. the camera or scanner. You need to design it to be robust, i.e. to handle various sizes, orientations, lighting conditions, etc. If it cannot handle these you need to define a calibration routine that is easy and fast. As part of the demo we will test the robustness via having the program classify "unknown" input.

After you have defined the problem you need to generate a test image database. For example, if you are identifying geometric shapes - circles, triangle, squares, rectangle, etc, collect via the camera or scanner (at least) 6 images of each class (the more the better). BEFORE you start algorithm development, separate the images into two sets - one training set and test set. Next, use the training set to develop your algorithm with CVIPtools.

The following guidelines will help:

• 1) use CVIPtools to explore algorithm development, using primarily the ANALYSIS window for these types of projects.
• 2) First, segmentation by using the Segmentation window and/or edge detection.
• 3) Next morphological filtering to reduce and solidify spatial objects (at bottom of Segmentation window).
• 4) Next, Features window to do feature extraction.
• 5) Examine the feature file to come up with an classification algorithm.
• 6) Code the algorithm you developed into your CVIPlab program. This is done by:

· a) finding the function name that corresponds to the CVIPtools function, see Chapter 12 and Appendix C in the book and the Help in CVIPtools.

· b) See the Help pages "CVIPtools C Functions" to see how the function is used in a C program (note html versions of the Help pages are also on the Web at http://www.ee.siue.edu/CVIPtools).

· c) do this for all functions you need.

· d) code the algorithm with these functions in your CVIPlab program.

· e) code your classification algorithm into your CVIPlab program.

• 7) Test your program on images you input from the camera.
• 8) Write a report on your results, and present a demo to the class. The project is graded based on what you did, so tell me what you learned, not just the "good" results.

Notes:

• Be sure to use real images (camera or scanner input), not computer generated images.
• Use a fixed set of images, at least 10, to develop the algorithm. Test it on a different set of at least 10 images. Use images with multiple objects. Tabulate or graph statistical results, compare various algorithms you have developed, or compare use of varying parameters.
• Use the HELP pages in CVIPtools if you have questions on using the software (especially Feature extraction), if you still have trouble talk to me (sumbaug@siue.edu, 650-2948).
• If you are not a real C programmer, I suggest you leave out step 6 and do more work in step 1. I encourage the use of CVIPtools, and welcome any feedback on the experience.
• If the problem gets too complex to get anywhere in the time allowed, extract a portion of the problem, or simplify.
• Try using the Utilities for common functions. Pay attention to the DATA TYPE of output images, this can be converted in Utilities/Convert. Note that images are always remapped to BYTE data type for display, but the original image is still in the IMAGE QUEUE. Utilities/Stats/Image Data Range will provide more information than the viewer about the actual data (the viewer will display minimum and maximum across all bands, real and imaginary data, while the Utility will display information for each band separately).