Composition modeling for photo retrieval through geometric image segmentation

摘要

A composition model is developed based on the image segmentation and the vanishing point of the scene. By integrating both photometric and geometric cues, better segmentation is provided. These cues are directly used to detect the dominant vanishing point in an image without extracting any line segments. Based on the composition model, a novel image retrieval system is developed which can retrieve images with similar compositions as the query image from a collection of images and provide feedback to photographers.

主权项

1. A method of analyzing a photographic image, comprising the steps of:
receiving an image at a computer processor operative to perform the following steps:
determining a vanishing point in the image,providing a segmentation which generates regions and boundaries between two of the adjacent regions, wherein a plurality of the regions is characterized by photometric cues and geometric cues, the geometric cues including the vanishing point,weighting the photometric cues and geometric cues,determining a weight for a boundary between two of the adjacent regions as a combination of the weighted photometric cues and weighted geometric cues between the two of the adjacent regions,executing a hierarchical image segmentation process on the image to obtain an image segmentation map that partitions the image into photometrically and geometrically consistent regions based on the determined weight for the boundaries, andmodeling the composition of the image based upon the image segmentation map and the vanishing point.