Bidirectional Matching Algorithm for Target Tracking Based on SIFT - Advanced Research on Computer Education, Simulation and Modeling - page 254

2 The Extracting of the SIFT Feature

2.1 The Building of the Scale-Space

To detect the keypoints in the scale-space effect, SIFT use the gauss function to build
the scale-space.

( ) ( ( ) ) ( ) ( ) ( ) D x , y , σ = G x , y , k σ − G ( x , y , σ ) × I x , y = L x , y , k σ − L x , y , σ

(1)

( )

G ( x , y , σ ) I x , y

In the last formula the is the gauss function, the is the image function [2].

2.2 Keypoint Detect

To find the extreme points of one image in the scale-space, the centre sampling point
will be compared with the eight points that are around it in the same scale and the
eighteen points that are in the up scale and the next scale, as shown in the Figure 1.

Fig. 1. keypoints detect [3]

2.3 Descriptor Representation

These samples are accumulated into orientation histograms summarizing the contents
over 4

× ×

2 descriptor array
computed from an 8x8 set of samples, whereas the experiments in this paper use a
4

4 subregions, as shown on the right. This figure shows a 2

× ×

4 8 = 128 element feature vector for each keypoint [4].

Fig. 2. The creation of a keypoint descriptor