Content-based image retrieval using integrated features and multi-subspace randomization and collaboration

In the Content-based image retrieval process, the semantic gap is a common challenge and the existing Bag-of-words approach is used to reduce the semantic gap. However, the computation complexity was the common problem that occurred during segmentation and the existing model underwent difficulty during the process of image retrieval. The retrieval performance influenced in exploring various sub-spaces for high dimensional data. The proposed approach performed integration of features such as local optimal oriented pattern, grey level co-occurrence matrix, and Alex net from convolution neural network with the multi-subspace randomization and Collaboration for the retrieval of semantic image. Firstly, the image segments the foreground objects from the background regions using Super Pixel-based Salience Segmentation. From the segmented region, the features present in the objects are extracted. The obtained segmented regions are used for integrating the features and will balance the data dimensions of each of the image pixels. The balanced subspace randomization schemee produces multiple partitions of features that are similar-sized random sub-spaces based on the Manhattan distance. The existing models utilized an ImageNet image repository Interval Type-2 Beta fuzzy near sets Corel datasets that obtained average precision of 78.34%. Similarly, fuzzy C-Means clustering and soft label support vector machine utilized both corel and VOC that obtained average precision of 91.45% and 90.72% better when compared to the proposed method that obtained Average precision of 94.36% and 96.03% for Corel and VOC datasets.