An Improved Image Descriptor for Image Classification and CBIR Applications

Recent technological advancements have significantly expanded multimedia complexity, opening up new research fields that rely on similar multimedia material retrieval for image classification and CBIR. Using image features from those images, such as color, composition, structure, and spatial configuration, rather than manually annotated textual keywords, content-based image retrieval (CBIR) searches for images in a massive dataset that are comparable to a given query image. In this methodology, a histogram of oriented gradients (HOG), Color Histogram along with HSV are used for extracting shape and color. The Gabor Wavelet Filter has additionally been applied to object and texture recognition. And an improved convolutional neural network (CNN) is used for similarity matching. In this project, the experiments run on image database having 19 categories and each category contains one hundred images.