Tri-Chrominance Texture Pattern: A New Feature Descriptor
Israel Jeena Jacob, K. G. Srinivasagan, D. Gomathi, P. Joyce Beryl Princess, P. Betty, P. Ebby Darney · International Journal of Engineering & Technology · 2018
Feature extraction plays a vital role in the information management system. This paper proposes Tri-Chrominance Texture Pattern (TCTP), a feature descriptor for extracting the features from images. This pattern helps to extract the inter-channel chrominance relationship, along with texture information of the image. The analysis were done in a natural image dataset, Corel database (DB1), pure colored texture database, Colored Brodaz Texture database (DB2) and a biometric dataset, Indian Face Image database (DB3). The proposed work outperforms the existing works in all the datasets. The analysis on DB1 shows significant improvement over the previous works like Local Binary Pattern (LBP) (78.64%/57.35%), Local Tetra Pattern (LTrP) (79.84%/56.8%) and Local Oppugnant Color Texture Pattern (LOCTP) (82.64%/58%) as 83.25%/58.2% in terms of Average Precision/ Average Recall. The analysis made in the Colored Brodaz database (DB1) shows the result of TCTP as improved from LBP (91.75%/75.18%), LTrP (91.64%/76%) and LOCTP (99.21%/89.38%) to (99.8%/93.47%). The Average Recognition Rate (ARR) of face recognition in DB3 database using the proposed work shows considerable improvement from LBP (78.2%), LTrP (91.9%) and LOCTP (89.1%) as 88.5%. The computational complexity of the proposed work is much lesser than LTrP and LOCTP.