Method, apparatus and system for determining a saliency map for an input image

摘要

A method of determining a saliency map (e.g., 900) for an input image (e.g., 600) is disclosed. Pre-determined data defining relative salience of image features is accessed. The input image (600) is decomposed according to the predetermined data to identify features of the input image (600) corresponding to the image features in the predetermined data. A portion of the predetermined data corresponding to a range of the identified image features is selected. A perceptual scale is determined using the selected portion of the predetermined data. The saliency map (900) for the input image using the determined perceptual scale.

主权项

1. A method of determining a saliency map for an input image, said method comprising:
accessing predetermined data defining relative salience of image features, the predetermined data representing saliency observations obtained using psychophysical experimentation, and the predetermined data relating to multiple image features that are correlated with human perceptions of saliency; decomposing the input image according to the predetermined data to identify image features of the input image that have corresponding image features in the predetermined data; obtaining a perceptual scale, for each of the identified image features of the input image that have corresponding image features in the predetermined data, by:
selecting a portion of the predetermined data corresponding to a range of feature data values associated with the respective identified image feature, anddetermining the perceptual scale for the respective identified image feature using the respective selected portion of the predetermined data, the perceptual scale for the respective identified image feature being based on probabilities determined using the respective selected portion of the predetermined data; and determining the saliency map for the input image using the determined perceptual scale or the determined perceptual scales.