Fast and Accurate Content Based Image Classification and Retrieval using Gaussian Hermite Moments applied to COIL 20 and COIL 100
Bollineni Sai Mohan, T. Krishna Chaitanya, Talari Tirupal · 2019
Image retrieval based on content is gaining popularity due to the rapid progress in the usage of image a data over mobile and internet. Fast and accurate retrieval of the images is always essential in particular when the bandwidths are scarce. Many CBIR algorithms are available in the literature for various databases. This paper proposes a CBIR algorithm for improving the retrieval accuracy from COIL 20 and COIL 100. Both the databases are basically shape based ones. Even though many shape based descriptors are available image moments play an important role in the description of the shape. Some moments which are extensively used are Hu's moments, Fourier Mellin Moments, Zernike Moments, Legendre Moments, Gaussian Hermite moments, Chebycheff moments. In this work, an algorithm using Gaussian Hermite moments and SVM is proposed. Medium Gaussian kernel is used in the SVM. A cross validation of 10 fold is used for the simulations for COIL 20 database. 25% hold out validation is used for COIL 100 database due to its large size. Performance metrics used are precision and recall in addition to the average retrieval efficiency. An average retrieval accuracy of 100% is obtained for COIL 20 database. For COIL 100, the maximum average retrieval accuracy is 99.5%. 100% precision and recall are obtained for both the databases. Results are obtained with few features, the proposed algorithm is fast compared to many algorithms. Proposed algorithm is compared with the state of the art algorithms available in the literature.