Point of feature extraction algorithm

Abstract: In photogrammetry, there are some of the more well-known feature extraction operator points, such as: Moravee operator, Forsmer operator and Hannah operator, etc. will be described Moravec operator and the basic principles of Forsmer operator, from the extraction point positioning accuracy and speed are two aspects to compare the two operators, and focus on analysis using Moravec operator to extract feature points analysis of the implementation process.

Keywords: feature extraction, point features, Moravec operator

Point is characterized by the most basic features of the image, it refers to those in two-dimensional gray-scale signal has a significant change in the direction of the point, such as corner points, dots, etc. can be applied to point features such as image registration and matching, the target description and identification, beam calculation, moving object tracking, recognition and stereo and many other areas of 3D modeling using point features for processing, can reduce the amount of data involved in the calculation, without compromising the importance of gray scale image information in the match computations can be greatly improved matching speed, and thus people's attention. extracted point feature called interest operator operator operator or favorable (interest Operator), namely the use of an algorithm to extract from the image of interest, conducive to the purpose of a point in the image analysis and computer vision field, according to different purposes and effective point selection feature extraction.


1 Moravec interest operator

Moravec in 1977 proposed the use of gray-scale variance extracted feature point operator. Moravee operator in the four main directions, with the most one minimum point as the gray variance feature points.

The first step, calculate the value of the pixel of interest IV (in terestv aIue).

The second step, given a threshold experience, the interest is greater than the threshold point (ie, interest value of the window center) as a candidate point threshold should be chosen to point in the candidate feature points including the need to , fork without too much of the principle of non-feature points.

The third step is selecting the candidate point in the extreme points as feature points.

In addition to the above method, you can also try to first use edge extraction method to extract the image edges, and then use this profile within the feature point extraction method to extract these feature points.


2 Forstner interest operator

Forstner operator is extracted from the image points (corners, dots, etc.) characteristic of a more effective operator. Foratner operator by calculating the gradient of each pixel and the pixel Robert (c, r) as the center of a window covariance matrix of gray in the images as small as possible and to find a point close to the circle as the feature points, it is by calculating the value of each image point of interest and the use of inhibition of local minimum method of extracting feature points.

Step 1: calculate each pixel of the Robert gradient,
Step Two: Calculate the gray-scale 1 × 1 window covariance matrix.

The third step: Calculate the interest value of q and w.

Step Four: Determine the candidate point.

Step five: Select the extreme points.


3 Moravec block diagram (Figure 1)

4 points based on Moravec operator feature extraction map
Grayscale renderings are as follows:
Operator characteristics can be seen to increase contrast of surface features of some of the better edge detection, edge detection and contrast the smaller less effective, which is the threshold and window size selection algorithm itself is determined. Share for free download http://www.hi138.com 5 Conclusion

Moravec operator is a point feature extraction operator operator in one of the classic, then a lot of point feature extraction operators are based on the improvement in its come to master Moravee operator theory and implementation methods for understanding the other point feature operator understanding and application of great benefit.

Links to free download http://www.hi138.com