Low-complexity method of converting image/video into 3D from 2D

摘要

A low-complexity method of converting 2D images/videos into 3D ones includes the steps of identifying whether each pixel in one of the frames is an edge feature point; locating at least two vanishing lines in the frame according to the edge feature point; categorizing the frame into the one of close-up photographic feature, of landscape feature, and of vanishing-area feature; if the frame is identified to have the vanishing-area feature or the landscape feature to generate a GDM; and apply a modificatory procedure to the GDM to generate a final depth information map; if the frame is identified to have the close-up photographic feature, distinguish between a foreground object and a background information in the frame and define the depth of field to generate the final depth information map.

主权项

1. A low-complexity method of converting 2D images/videos into 3D ones for converting frames of an image/video one by one, comprising steps of:
a) identifying whether each pixel in a frame is an edge feature point; b) carrying out calculation for each pixel, which is the edge feature point and centered, by locating other pixels, which are also the edge feature points, around the centered pixel, saving data of straight lines connected between the adjacent pixels and the centered pixel separately, prioritizing the data of the straight lines according to the number that each of the straight lines passes through the edge feature points, and defining two straight lines passing through the most edge feature points in number as vanishing lines; c) identifying whether the density of the edge feature points in the frame is larger than a density threshold and whether the number of the vanishing lines in the frame is larger than a number threshold; if both are larger than the respective thresholds, it will be identified that the frame is of close-up photographic feature; and then calculating the ratio of sky to faraway mountain by color; if the ratio is larger than a landscape threshold, it will be identified that the frame is of landscape feature; if the frame is identified to be of neither the close-up photographic feature nor the landscape feature, it will be identified that the frame is of vanishing-area feature; d) locating the vanishing area of the frame by a vanishing-area detective technology, which is to locate the area where vanishing points that are crossing points of all vanishing lines are the densest in the frame and to define the area as a vanishing area, if the frame is identified to be of the vanishing-area feature; if the vanishing area is located outside the frame, set the border of the frame, which matches the tendency of the vanishing lines the most, as the vanishing area; if the frame is identified to be of the landscape feature, it will be defined that the vanishing area is located at the upmost side of the frame; next, generating a gradient depth map (GDM) based on the distance between each pixel and the vanishing area in the frame; and carrying out a modificatory procedure of the GDM to further generate a final depth information map; and e) applying pixel-based low-pass filtering to the frame, if the frame in the step c) is identified to be of close-up photographic feature, to distinguish between foreground object and a background information, defining that the foreground object is of the closest depth of field and the background information is of the farthest depth of field, and further generating a final depth information map.