The Extraction And Application Of Shape Invariants

Shape is an advanced visual feature exists in the image, and it can be used to represent the structure and attitude of the object, which has the characteristics of texture information not available. The shape feature extraction is the foundation of shape analysis and recognition, especially in a wide variety of geometric transformations (such as translation, rotation, affine, projective invariant features,etc). It has an important value and already been widely used in target recognition, image retrieval, image registration, product testing, biomedical engineering and other scientific research and engineering technology field.This paper is based on the algebra and geometry invariants, shape invariant features are constructed by the corresponding distance constraint, angle constraint, position constraint, and so on. We mainly focus on the geometric features with affine and projective properties, and construct shape descriptors for complex shape. The correlation method is also extended to the recognition of3-D motion curve. In this paper, the main research contents are as follows:Firstly, according to the analysis and comparison of the existing shape feature descriptors, complexity of the shape feature descriptor and the matching efficiency are the two factors of the process. In this paper,a piecewise constraint algorithm based on measurement information is proposed. Compared with the traditional shape features of shape context, the method in this paper has less calculation complexity, lower dimensionality. And it is more stable to local deformation, and the method is applied to the Euclidean distance, triangle radius,angle correlation etc., which can describe the relationship between sampling points.Secondly, by the comparison of the existing affine invariant features extraction method of the shapes, and Characteristic Ratio can describe the position relation between the multiple collinear points. Based on the invariant, the relationship between the internal structure and the outer contour of the shape can be obtained. The invariant feature descriptor for complex planar shape is constructed. By comparing with the existing affine invariant feature descriptors, the experiments results show the stability and effectiveness of this method. Further more, our method also has a good distinction to similar shapes and shapes with noise.Thirdly, this paper analyzes the shape feature extraction method and the advantage of the traditional feature extraction method based on projective invariants. A hierarchical projective invariant descriptor is proposed, by the combination of projective invariants and context method. The contour points are sampled by the coarse to fine manner. The hierarchical description can describe the shape from overall to local, which has both projective invariant property and distinguishable ability.Finally, in this paper the trace of the key points of human body is treated as a spatial curve, and the projective invariant characteristic number is used to describe the3D spatial shapes, which can resolve the recognition problem of multi-angle of action sequences. This paper presents two methods named time sequence and spatial sequence based on monitoring key points. Experimental results show that the proposed method does not need training assisted like most multi-view action recognition methods, and it has higher flexibility in practical.