Novel Content Based Image Retrieval—Features of Correlated Visual Textons and MQLPP Descriptor

Content-based image retrieval (CBIR) has made significant advancements and continues to evolve with the rapid development of computer vision and artificial intelligence technologies. Feature plays vital role in CBIR systems since its main components are color, texture, and shape. Generation of the feature descriptor is a crucial part in CBIR due to its efficient exploration of image characteristics. Feature descriptors have the ability to reduce the feature dimension and enhance the image retrieval rate in CBIR. Hence, in this paper, a novel CBIR system is proposed named ‘Novel CBIR method based on features of correlated Visual Textons and MQLPP Descriptor (CBIR_VTMD)’, which aims high retrieval performance via finer feature extraction. The main contribution in the proposed method is a new texture descriptor namely ‘Multi-channel Quantized Local Penta Pattern based image descriptor (MQLPP)’ which expresses a new perspective to the enhancement of foreground object feature in an image. One of the significant traits of the MQLPP descriptor is the ‘utilization of innovative penta pattern using multi channels’. Proposed CBIR_VTMD method is experimented using well-known databases such as COREL-10K, CTMOS, ESAT, VISTEX, INDOOR, and DERMO. The experimental results reveal that the proposed CBIR_VTMD method enhances up to 10.7497% retrieval accuracy when compared to the state-of-the-arts method. The proposed method acts as the generic CBIR which efficaciously delivers retrieval results for the domains that uses natural, medical, remote sensing, and texture images. Besides, the proposed CBIR_VTMD framework works better in both online and offline real-time applications.