Application And Research On2D/3D Conversion Algorithm Of Images And Videos

There’s been a long history of research on the production and display technology of stereopsis. With recent development of3D film industry, the market has requested larger3D source and higher content quality.2D/3D conversion technology becomes an important mean to fulfill those requirements. With regard to static single-frame image, we in this article propose a complete set of methods to obtain depth map through the segregation and fusion of background space and foreground objects, using linear perspective depth cue.We systematically comb the stereovision principle and elaborate depth cues from perspectives of physiology and psychology, with an emphasis on binocular parallax and its relevance with stereoscopic shoot and display technology. A survey on the upstream and downstream technology of2D/3D conversion has been done.As for2D images with perspective characteristics, we binarize it with Canny edge detection algorithm. The resulted gray scale image is then used as the input of Hough transform, leading to a collection of lines, from which we filter to get vanishing lines and point. Main structure lines are extracted from vanishing lines through image semantic classification. Background space depth map is obtained through the position of vanishing point and main structure lines.We segment the original image interactively with GrabCut, resulting in foreground objects with which we isolate masks. Depth is then assigned according to masks’ position relation with background space. The result of foreground depth assignment is merged with background space depth and the final depth map is generated. DIBR is used to render anaglyph.