6-4 Haar Features 1

In fact, the characteristic is the pixel point of an area in an image, and the result is obtained after some kind of arithmetic. So the feature of the image is that it is a result of a pixel after it has been calculated. The result can be a specific value, a vector, or a multidimensional element. So, the nature of the feature is actually the result of the operation of the pixel.

The previous section gives you a matrix operation. How do you use features to differentiate your goals? The threshold value decision. If this trait is greater than a certain threshold, then we can think of it as a target. If it is less than a certain threshold, then it is considered a sub-goal. So this is one of the simplest sentencing situations.

We get this verdict clearly through machine learning. After the verdict is clearly obtained, we then conduct the threshold judgment according to the characteristics of the calculation.

For example, our template size is a 10*10 matrix, which covers a total of 100 pixels, so the black part has a total of 50 pixels, and the white part covers an additional 50 pixels. The current black-and-white template is placed anywhere on the entire image, and the sum of the pixels in the white area (that is, the sum of the 50 pixels covered by the white area) is subtracted from any position, and we are given the characteristics of the 50 pixels of the black pixel coverage area. A feature is actually a arithmetic of pixels within a region. The entire area refers to 100 pixels covered in black and white.

The first formula and the second one must be the same.

# Haar 1 What is haar? feature = pixel operation-"result (concrete value vector matrix multidimensional) # 2 How do I use features to differentiate goals? Threshold Judgments # 3 getting a verdict? Machine learning # 1 Characteristics 2 verdict 3 Get verdict # Formula derivation 1-2 # feature = entire area * Weight 1 + Black * weight 2 = (Black + White) * 1 + Black * ( -2) = # = black + White-2 black = white-Black

6-4 Haar Features 1