期刊名称：International Journal of Software Engineering and Its Applications
印刷版ISSN：1738-9984
出版年度：2014
卷号：8
期号：3
页码：189-202
DOI：10.14257/ijseia.2014.8.3.17
出版社：SERSC
摘要：In order to manage and retrieve images from a large database based on multi-media content features, the features should have enough differentiation and be invariant to expansion, movement, and rotation. Content features include information on colors, shapes, textures, and movement. In particular, as human eyesight is sensitive to shape information as well as illumination or color in early judgments on objects, there have been a number of studies on the detection of objects. Therefore, this study proposed a new algorithm where noise invariability is added to the existing standard Hough transform which is applied for recognition of shape.For the order of movement, first, feature points are extracted through the standard Hough transform from a query image. Second, the numbers of lines which pass through each of the feature points are voted on and the voting numbers are rearranged according to size. Third, labels for the voting numbers are clustered and normalized to compose a feature table. In the simulation of the image retrieval system using a voting numberbased rearrangement Hough transform, a variety of images with different sizes were analyzed. As a result of the analysis, it was discovered that the weak point was that this transform was sensitive to rotation, expansion, and reduction. If the algorithm used in the standard Hough transform was complemented, it could improve recall and precision greatly in image retrieval.
