An Improved Method for Retrieving Fashion Images Based on Parsing the Image Contents

The field of fashion image retrieval is an active research area. In this paper, we rely on CBIR systems to retrieve fashion images. These systems depend on analyzing the contents of images in terms of features. We propose an improved fashion image retrieval method based on shape and local color features. In order to extract the shape feature, we have improved the VGG-16 model. To extract the local color feature, we use both the SRD and DCD from the MPEG-7 methods to remove the background and extract the dominant color. We also use the indexing method to quickly retrieve images. In order to retrieve the images that are most similar to the query image, we used the dissimilarity measure. The results showed that the proposed method determines the shape of a piece of clothing by 100/100 and that the percentage of time it takes (0.27%) to retrieve images is compared to a sequential search and is more accurate than a sequential search.