Visual invariant theory is eighteen eighties at the end of the invariant theory is introduced to a research direction in the field of computer vision research. Visual invariants are those changes in scale, moving, rotating, radiation, the perspective changes remain invariant feature. The SIFT algorithm is proposed in this study is a feature extraction algorithm based on local invariant. SIFT (Scale Invariant Feature Transform) algorithm proposed by D.G.Lowe in 1999, and in 2004 made a perfect and summary. SIFT feature points for image rotation and scale invariance to illumination changes, and from the perspective of 3D can also maintain a certain stability, but due to being localized well in three-dimensional space and frequency domain, so it reduces the possibility of noise. This paper discusses a simple method of SIFT operator, operator is simplified greatly reduces the dimensions of feature point descriptor is compared with the original operator. At the same time, SIFT operators simplified invariance of good, in the 3D perspective changes, illumination changes also had good stability, and the time complexity is greatly reduced compared with the SIFT algorithm, more suitable for mass data searching and matching. The main contents of this paper are as follows: (1) the classification method of transformation matrix and the image features of the computer vision and visual feature invariant theory, the feature extraction, and finally summarizes the main research content and organization (2) introduced the SIFT operator, the multi-scale space and Gauss differential method, described the process of SIFT operator extremum extraction, removal of unstable point, feature point descriptor generation, this operator remains committed to scaling, rotation, affine transformation, the light also has a certain invariance. (3) to improve the SIFT operator, the dimensionality reduction using a simplified rectangular SIFT feature vector before concentric circle window replacement.
