A corner is a point that changes in multiple directions at the same time. It is more unique than an edge. Corner detection is widely used in many scenarios, such as fingerprint feature extraction in fingerprint detection and sparse optical flow computing.
The basic idea of Harris corner's corner points is to observe through a small window, so it is easy to identify the corner points. In these directions, the brightness distribution of the window varies greatly if you want to move the window in any direction. Points in an image can be divided into three types: "flat", "edge", "Corner", and Harris.AlgorithmIs to extract the "corner ".
The following uses a 2D grayscale image to give a mathematical derivation of the Harris corner detection. The image is recorded as I, taking into account the image blocks within the range of (u, v), and translating (X, y) after the image block, the two weighted SSDs are recorded as s:
Approximate I (U + X, V + Y) to Taylor.
The above formula
Here, IX and Iy are the partial direction of I in the X and Y directions respectively. The preceding formula can be written as follows:
Where a is
<> Indicates that the average value is obtained in (u, v) blocks. Generally, the guass window is used.
S greatly changes the translation of all directions. Therefore, by analyzing the feature value of A, you can determine whether a point is a corner point:
1. If and, there are no features of interest in (x, y) pixels.
2. If it is a relatively large positive value, it is the edge point.
3. If the sum is a relatively large positive value, it is the corner point.
In practical applications, the computational complexity of directly solving feature values is relatively high. Therefore, Harris and others proposed to use the following Mc function for measurement, where k-k is an adjustable sensitivity parameter, experience Value 0.04-0.15 is generally used:
Therefore, the algorithm does not need to directly calculate the feature root, but to convert the problem into the determination value and trace solution.
The lab result is