SSIFT: An Improved SIFT Descriptor for Chinese Character Recognition in Complex Images

Text is a vital feature in applications of computer vision. Traditional Chinese character recognition techniques are mainly based on optical character recognition (OCR), however, they can't obtain satisfactory results from images affected by complex circumstance, such as different viewpoint, scale changes, addition of noise and complex background. To solve these problems, inspired by SIFT descriptor, we innovatively propose an improved feature descriptor, SSIFT (shape-SIFT), combined SIFT with relatively global shape descriptor towards recognition of segmented Chinese character. The experimental results on different datasets acquired under complex circumstances, indicate that the mixed descriptor distinguish similar local parts of various characters well. It obtains comparable results with SIFT and even outperforms SIFT in certain aspects. This novel Chinese character descriptor is illustrated to be feasible and effective.