A Novel Image Retrieval Technique based on Gabor Function, Local Tetra Pattern and ASMC

CBIR alone won’t give perfect retrieval results due to semantic gap. To overcome the problem of semantic gap in CBIR, more than one Semantic Content Based Image Retrieval techniques are required which is known as Hybrid Classification System. Hence the proposed approach uses multiple machine learning techniques with combination of different classifiers like supervised and unsupervised, soft classifiers, spectral contextual classifiers. Remotely Sensed Image Retrieval System (RSIR) has to identify and retrieve similar images based on query image, to do so we need to extract feature of image in order to compare query Image and database image. The proposed approach is a combination of two Phases. First Phase involves feature extraction by Texture Feature with the help of Gabor Function and Spectral Distribution using Advanced Split and Merge Clustering whereas second Phase identifies the Local Pattern of retrieved images in Phase-I. The performance of the proposed approach is measured in terms of Precision, Recall and f-measure. Statistical analysis of the proposed hybrid approach in Phase-I (Texture and Spectral Distribution) shows that precision, recall and f-measure is get improved, on an average by 19.46%, 8.84%, 14.46% respectively when get compared with CBIR (Texture). Phase-I and Phase –II comparison in term of f-measure is increased up to 96.95%. Hence the hybrid approach gives more accurate result as compare to individual approach General Terms Image Retrieval for dataset of satellite imagery