Opencv LBP features

One 、 brief introduction

Opencv And LBP features （ Algorithm ） - My pot - Blog Garden

LBP（Local Binary Pattern）, namely Local binary mode , It is a nonparametric algorithm that describes the local features of the gray relationship between image feature pixels and each pixel , It is also an efficient texture description algorithm . For a pixel with a radius r Draw a circle , Take... From the circle K A little bit （ It's usually 8）, this K The value of a point （ The pixel value greater than the center point is 1, Otherwise 0） form K Bit binary number . This is the local binary pattern , What is used in practice is LBP Square statistical graph of characteristic spectrum . In the old version of Opencv in , Use CvHaarClassifierCascade function , Only support Harr features . The new version uses CascadeClassifier class , It can also support LBP features

advantage ：

1, Rotation invariance （ The local binary loop shifts left or right, and its representation remains unchanged ）

2, To some extent, the problem of light change is eliminated

3, Texture feature dimension is low , Fast calculation

shortcoming ：

1, When the light changes unevenly , The size relationship between pixels is destroyed , Corresponding LBP The operator changes

2, By introducing the definition of rotation invariance , send LBP The operator is more robust . But it also makes the LBP The operator lost direction information （ For example, move the local binary left or right , The result is the same , But the image is different ）

Two 、 Gray scale invariance

  The central gray value of the texture unit （ threshold ） by 85, rest 8 The adjacent gray values are ：90、74、63、134、91、83、99、103、90. Calculate according to the theorem of threshold calculation , After the threshold calculation , According to the direction indicated by the arrow in the figure, a local binary pattern can be obtained , namely ：LBP Pattern =1101001, And LBP=217,. Through the above LBP The calculation example of the algorithm can be seen LBP The value is only determined by the gray value of the central pixel and the gray value of eight adjacent pixels ,LBP The feature represents the feature of its central pixel , The calculation of its value also needs to be completed with the surrounding pixels , therefore ,LBP Features have a lot to do with their surroundings , It contains image points , It also contains edge and local feature distribution information .

3、 ... and 、 Rotation invariance LBP Algorithm

Through the introduction above , It's not hard to see 1LBP The algorithm is gray invariant , stay 2002 Of the paper in , In addition to grayscale invariance , The author is for LBP Rotation invariance and equivalent mode are introduced , How does it work ？
We know from the above introduction that , For a selected P,LBP Algorithm will generate 2^P Different output values , such as P=8 when , The value is 256 individual , this 256 Each of the three modes rotates around the center , Then in the process of many rotations , There are many different results , And the mode with the smallest median of these results is to choose the constant binary mode . such as ：

  What is illustrated in the above figure is , This one in the picture 8 Patterns , All have the same rotation invariant mode , The decimal system is 15 The value of , namely 00001111. So right. 256 Both modes do this rotation , Get the smallest number as the rotation invariant mode of this mode , There are rotation invariant modes 36 Kind of