TLD algorithm Feature Selection

TLDAlgorithmThe first task was to select a feature. I found the answer in one of the documents.

The author just made a comparison, the method used is a stupid method. How can I find an optimization? This can be used as a good idea. I am mainly having limited time. If anyone has any idea or perseverance, I will do this. hope to tell me!

Understanding this feature selection requires a certain mathematical foundation,

1 Bayesian Probability Model

2 Mean-Shift Algorithm for tracking

3. Image Histogram

4 Fisher Discriminant

If you know more about these aspects, I think you can understand what I'm talking about.

The author mainly optimizes an expression.

F = W1 * r + W2 * g + W3 * B

Generally, image processing uses the default value for this expression. However, if you select the weights for image tracking, the background and items of the image can be distinguished!

(Because I am too lazy, I want to give specific instructions on how to get these values. I hope you will pay attention to your blog in a few days and leave a lot of messages to stimulate my enthusiasm. Heheh)

PS:

I recommend a few good blog posts on the above mathematical knowledge.

1. http://mindhacks.cn/2008/09/21/the-magical-bayesian-method/ (Bayesian)

2. http://download.csdn.net/detail/muzi198783/4135834 (mean-shift, Fisher)

3. Just look at the digital image processing book about the image histogram.