Image segmentation using color and depth information

摘要

Image segmentation utilizing 3D image data. A plurality of pixels of an image frame may be segmented based at least on a function of pixel color and a pixel depth over the spatial positions within the image frame. A graph-cut technique may be utilized to optimize a data cost and smoothness cost in which at least the data cost function includes a component that is a dependent on a depth associated with a given pixel in the frame. In further embodiments, both the data cost and smoothness functions are dependent on a color and a depth associated with each pixel. Components of at least the data cost function may be weighted for each pixel to arrive at most likely segments. Segmentation may be further predicated on a pre-segmentation label assigned based at least on a 3D spatial position clusters.

主权项

1. An apparatus, comprising:
an input to receive image data including at least color information and depth information for each of a plurality of pixels of an image frame; a computer processor coupled to the input, the processor to:
assign pre-segmentation labels to pixels corresponding to three-dimensional (3D) spatial position clusters that satisfy a minimum cluster size;assign pre-segmentation labels to pixels corresponding to color clusters determined for those pixels that failed to satisfy the minimum 3D spatial position cluster size;determine a data cost associated with assigning a segmentation label to each of the pixels based on each pixel's depth, color, pre-segmentation label, and position relative to the nearest pixel assigned the pre-segmentation label equal to the segmentation label;determine a smoothness cost of assigning different labels to neighboring pixels as a function of the pixel depth and the pixel color relative to one or more neighboring pixels;select the segmentation label for each pixel based on a graph-cut that optimizes the data cost and smoothness cost; andoutput image segmentation data that maps the selected segmentation labels to the corresponding pixels of the image frame; and an electronic memory coupled to the processor to store the image segmentation data.