|
Results
After synthesizing all of the individual parts of the Playing Card Image
Recognition System (PCIRS), we ran a four types of tests:
- Single Card
- Single Card - Rotated
- Multiple Cards - No Overlap
- Multiple Cards - Overlap
Here are the results:
Results of testing
In general, our algorithms proved very successful. We were actually able to
"make it work." We starting with an idea and implemented it which is an amazing feeling.
With single cards, the system worked 100% of the time. With the multiple card images,
the resulting outputs decreased. This was caused by limitation in the Roberts' Cross
filtering. Ultimately, however, our group was quite pleased with the project.
Of course, there are some improvements that the
nPhaze Boys would like to see implemented. No system is quite perfect.
- Instead of using Roberts' edge detection in locating the corners,
create an erode/dilate function to increase corner detection accuracy.
This would help in the multiple cards - Overlap category.
- Decrease runtime in the scanning algorithm. Make it more "realtime."
- Add capability for multiple cards, overlapped, with variation in
rotation to cover the most general case.
MATLAB Code
Here are our MATLAB code files:
Edge/Corner Detection
- calcspot.m - sorts the potential detections
of the card corners, takes the median of the row and column values of each
hotspot, stores them as the valid corners (upper left point of the corner).
- cutout.m - cuts out the corners of the image. Used
to check visually that the corners were correctly found.
- findspot.m - wrapper function which performs
the whole edge/corner detection.
- scanit.m - rasterizes through the image with the
scanning window, summing white pixels, saving potential detection.
- showdetect.m - displays a graph of all the
potential detections. Used for debugging.
- showpic.m - displays any matrix with grayscale
colormap and square axes.
Correlation and Point of Control
- identifyMultipleCards.m - manager
function of the scanning and correlation process. The nachos grande so to
speak.
- identifySingleCard.m - takes in
the coordinates of a corner and returns the card value
(in integer form).
- library.mat - contains the library reference
images used in the correlation.
Rotation
- cropit.m - cuts off the excess black space
created by rotation, used to keep images a manageable size.
- findang.m - performs the gruntwork of
finding the angle from the Radon transform for rot8.m
- findleft.m - utilized by scan.m, slides down
the edge of the card to find the leftmost edge, give the top corner
as the starting point.
- rot8.m - takes in the image and finds the angle
by which it has been rotated using a Radon transform.
- rotback.m - takes in the image and magic
angle, returns the upright image.
- rotcard.m - wrapper function that calls
the other functions with the correct nesting.
- scan.m - finds the upper and left bounds of
a rotated card so that it may be cropped from the rest of the image.
Demos
Here are three demo M-files that can be run to see
the Playing Card Image Recognition System (PCIRS) in action. Please type UNLIMIT
at the prompt before entering MATLAB to release memory requirements. Also, please
be patient. Run time reaches up near a minute for the large image files.
- demo1.m - Single Rotated card
- demo2.m - Multiple cards - diagonal overlap
- demo3.m - Multiple cards - straight overlap
 Next: Page 4
Previous: Page 2
|
|
