Car Color Recognition Algorithms

Hello, I'm a final year student and now I'm dealing with a project which is about "Car Color Recognition". I've read some papers and tried algorithms. Some of them can be listed as;

- Used different color spaces like rgb, hsv, lab. Then calculated mean or median values of pixels to get vector for svm. While doing this I did not use illumination factors to avoid problems in this manner.

- Created color histogram on hsv-space and used maximum numbered bins then used svm again to classify them.

- I divided image into rectangles and weighted these rectangles differently to focus on the car which is at the middle of the image in general.

- I used expectation maximization for the previously mentioned vectors.

However, still I am getting results lower than my expectations. Images can vary a lot because of the illumination change. The process is stucked and I need some advices to move on. I really hope someone could help me in this process.