Design of Image Retrieval Descriptor Based on the Fusion of Colour and Texture Feature Descriptors
Himani Chugh, Meenu Garg, Sheifali Gupta · 2023
In the modern era, image retrieval research is quickly moving into a data-driven phase. Various methods can be applied for the detection of different features present in the image. Large picture datasets may now be created thanks to advances in image collection and data storage. In this case, it is essential to create the right information systems to effectively handle these collections. The primary focus of this chapter is the recovery of images utilizing their colour and texture attributes. A well-structured computer-assisted retrieval strategy that can employ a hybrid mix of an image’s texture and colour data is needed for this purpose. This study proposes a novel framework for image retrieval that increases retrieval precision by combining colour and texture features in a hybrid way. The colour difference histogram (CDH) descriptor is used to describe the colour features, while the microstructure descriptor is used to describe the texture features (MSD). The hue and saturation value (HSV) feature and the gray-level co-occurrence matrix are determined inside the CDH and MSD descriptors, respectively, to separate the distinct image features. A threshold value for Euclidean distance is established to retrieve images by comparing the concatenated properties of database images with the query image. The best outcomes are obtained when the threshold is set at 50% of the maximum Euclidean distance. Other performance metrics, such as precision, recall, and F-measure, are also used to evaluate the system’s efficacy, with the corresponding results at 0.82, 0.92, and 0.87, respectively; these values outperform individual feature descriptors.