Fashion cloth image categorization and retrieval with enhanced intensity using SURF and CNN approach

Purpose The attraction of online shopping has raised the demand for customized image searches, mainly in the fashion industry. Daily updates in this industry increase the size of the clothing database at a rapid rate. Hence, it is crucial to design an efficient and fast image retrieval system owing to the short-listing of images depending upon various parameters such as color, pattern, material used, style, etc. Design/methodology/approach This manuscript introduces an improved algorithm for the retrieval of images. The inherited quality of images is first enhanced through intensity modification and morphological operations achieved with the help of a light adjustment algorithm, followed by the speeded up robust feature (SURF) extraction and convolutional neural networks (CNN). Findings The results are validated under three performance parameters (precision, recall and accuracy) on a DeepFashion dataset. The proposed approach helps to extract the most relevant images from a larger dataset based on scores conferred by multiple cloth features to meet the demands of real-world applications. The efficiency of the proposed work is deduced from its effectiveness in comparison to existing works, as measured by performance parameters including precision, recall and F1 score. Further, it is also evaluated against other recent techniques on the basis of performance metrics. Originality/value The presented work is particularly advantageous in the fashion industry for creating precise categorization and retrieving visually appealing photographs from a diverse library based on different designs, patterns and fashion trends. The proposed approach is quite better than the other existing ML/DL-based approaches for image retrieval and classification. This further reflects a significant improvement in customized image retrieval in the field of the fashion industry.