Study On Seal Matching And Identification

With study of mature algorithms at home and abroad, paper presents the stepwise refinement based on template matching of the square seal registration algorithm and SIFT based on squares which are formed by segmenting seal.Firstly, Template matching stepwise refinement of the square seal registration algorithm roughly matches seal by math geometry computing.But angle of registration error is more than 2°. Secondly, the paper computes mutuality of seal by circumvolving the template. When the mutuality is max, angle which template circumrotates is that the seal need circumvolve. Angle of registration error is less than 2°after this Step. Finally, by the method of coordinate projection, the paper can exactly match the seal. Experiments statistics show that this arithmetic can reach request of exact matching.SSIFT arithmetic is based on cirques squares which are formed by segmenting seal. Firstly, the paper analyses SIFT advanced by David Glowe. Many scholars have testified that the character vectors of SIFT are invariable to eddy, move and affine change. But SIFT for seal identification has lots of time by experiments. SSIFT excludes partially redundant point by Sobel arithmetic operators and samples the seal again in polar coordinates. It uses Sobel arithmetic operators to eliminate the points out of edge. Then we disperse the seal into some small cirques. So character vectors matching are processed on the parallelism cirques. Experiment statistics show that SIFT algorithm needs to deal with more than 2000 feature points and time-consuming is about 40s.but SSIFT only needs to deal with about 1000 feature points and time-consuming is about 20s. Identification accuracy of SSIFT algorithm is less than SIFT algorithm.But it is still more than 93% .Time-consuming of SSIFT algorithm is half of SIFT algorithm.