Distribution-Based Dissimilarity Measures
In order for an image retrieval system to find images that are visually similar to the given query, it should have both a proper representation of the images visual features and a measure that can determine how similar or dissimilar the different images are from the query. Assuming that no textual captions or other manual annotations of the images are given, the features that can be used are descriptions of the image content, such as color [97, 62, 96, 91, 4], texture [25, 62, 6, 69, 56, 4], and shape [42, 62, 31, 44]. These features usually vary substantially over an image, both because of inherent variations in surface appearance and as a result of changes in illumination, shading, shadowing, foreshortening, etc. Thus, the appearance of a region is better described by the distribution of features, rather than by individual feature vectors.
KeywordsImage Retrieval Dissimilarity Measure Partial Match Image Retrieval System Histogram Intersection
Unable to display preview. Download preview PDF.