Freak Feature Extraction algorithm

Brief introduction

FREAK algorithm is a feature extraction algorithm proposed in the 2012 CVPR freak:fast Retina KeyPoint, and is also a binary feature description operator.

It is very similar to the brisk algorithm, the personal feel is in the brisk algorithm improvement, about brisk algorithm, see the previous blog: Brisk feature extraction algorithm. Freak still has scale invariance, rotational invariance, noise robustness and so on.

Freak algorithm

Sampling mode

In the brisk algorithm, the sampling mode is the uniform sampling mode (sampling at the same round fine interval); In the Freak algorithm, the sampling mode is changed, and it takes a sampling model which is closer to the retina receiving image information of the human eye. The image shows the human retinal topological structure, and the fovea region mainly deals with the high-progress images, and the para mainly deals with the low-precision image information.

It can be seen that the structure is composed of many different sizes and overlapping circles, the most central point is the feature point, the other center is the sampling point, the distance between the sampling point and the feature point, the larger the radius of the sampling point circle, also indicates that the circle of the Gaussian function radius greater

Feature description

F represents the binary descriptor, PA is the sample point pair (as with brisk), and n represents the desired binary encoding length.

Represents the pixel value of the sample point to the previous sample point in the PA, which, in the same vein, represents the pixel value of the latter sample point.

Of course, after the binary descriptor of the feature point is obtained, the feature extraction is completed. But Freak also proposed that the obtained Nbit binary descriptors to be screened, hoping to get a better, more recognizable descriptors, that is, from the N to remove some bad descriptors.

Reference documents

1, Freak:fast Retina keypoint[j],cvpr,2012.

2. Research and implementation of image Mosaic Fusion technology based on feature [d],2014.

3. Research of registration algorithm for new surface mount components [d],2013.

4. Research on human face Recognition algorithm based on feature point description [d],2013.

5. Research on identification method of occlusion workpiece based on feature matching [d],2014.

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

Freak Feature Extraction algorithm