IMPROVED METHODOLOGIES FOR 2D TO 3D STEREOSCOPIC VIDEO CONVERSION

Abstract

The proposed model utilizes the depth level of y dimensional axis and coordinates for 3D conversion of foreground objects. Histogram equalization is randomly applied for separating the foreground objects from the background. Each pixel at top most depth level of the foreground object is truncated and depth cue is calibrated for better perception using binocular vision methodology along with Anaglyph which is modeled on CYAN to yield better results. newline Key frame is a drawing that defines the starting and ending points of any smooth transition. The depth information about key frames is extracted from the 2D images. Then both foreground and background objects are extracted using background subtraction algorithm. From the generated region of interest, both forward and backward motion pixels are extracted in the form of vectors. Gabor filter is applied to decompose high pass luminance. The proposed model is experimentally tested on both right and left view and the result shows that the presented model has got an edge over the traditional model. The performance is achieved using CRYENGINE gaming engine with NVIDIA graphical support for increasing the rendering speed of frame Buffer. Experimental results show that 3D recognition rate of a single image in the 2D video sequence is up to 78% and display rate of the particular sequence are at 29 frames per second. Experimental results have been included and 3D display rate for the converted sequence is up to 92% accuracy and frame rates be 32 frames per second for FULL HD (1080 p).