发明名称 Method for in-image periodic noise pixel inpainting
摘要 A method for in-image periodic noise pixel inpainting is provided. It is determined whether a current frame includes periodic noise pixels, and locations of periodic noise pixels are identified. Non-periodic-noise pixels in a reference frame are utilized to inpaint the periodic noise pixels in the current frame.
申请公布号 US9224052(B2) 申请公布日期 2015.12.29
申请号 US201314064706 申请日期 2013.10.28
申请人 Industrial Technology Research Institute 发明人 Lin Che-Tsung;Lin Yu-Chen
分类号 G06K9/40;G06K9/00;G06T5/00;G06T5/50 主分类号 G06K9/40
代理机构 Muncy, Geissler, Olds & Lowe, P.C. 代理人 Muncy, Geissler, Olds & Lowe, P.C.
主权项 1. A method for in-image periodic noise pixel inpainting, comprising: determining whether a current frame comprises a periodic noise pixel group, and identifying locations of pixels of the periodic noise pixel group in the current frame if the current frame comprises the periodic noise pixel group; matching a plurality of non-periodic-noise pixels in the current frame with a plurality of non-periodic-noise pixels in a reference frame to obtain a pixel relationship between the current frame and the reference frame; and selecting corresponding pixels from the reference frame to inpaint the periodic noise pixel group in the current frame according to the locations of the pixels of the periodic noise pixel group in the current frame and the pixel relationship; wherein the pixel relationship between the current frame and the reference frame is modeled by an affine transformation matrix, a projective transformation matrix, or a transformation matrix or a transformation formula which describes the pixel relationship between two frames; a plurality of non-periodic-noise feature correspondences in the reference frame, a plurality of non-periodic-noise feature correspondences in the current frame, and a pixel relationship between the non-periodic-noise feature points in the reference frame and the non-periodic-noise feature points in the current frame are identified by a scale-invariant feature transform (SIFT) algorithm or a speeded up robust features (SURF) algorithm; and a plurality of feature correspondences of the reference frame and the current frame is substituted by a random sample consensus (RANSAS) algorithm or a least-squares method to obtain the affine transformation matrix.
地址 Hsinchu TW