Abstract
In this thesis, a new content-based retrieval method for lung CT images is proposed. Content-based image retrieval have been used in many domains. However, it is recognized that content-based image retrieval from an image database cannot be done completely automatically. In our approaches, delineating the abnormal parts in lung CT images is done by human. After the regions have been marked, we have to extract the features of these regions and the most significant feature is the texture of the region. For extracting the textures of abnormal lung CT efficiently, we use some DCT coefficients that represent gray level variations of image and the spatial relations between pixels. These DCT coefficients are used as feature vectors that can identify different textures. Then, we use these DCT coefficients through Kohonen self-organizing network to cluster different textures. In addition to texture, theses abnormal regions are characterized by position and size. We use hierarchical similarity measure for retrieval. At first, check which texture category the query image in. Then, local search within the same category has been done using position and size.