An Evaluative Model for Discriminative Dictionary Learning with Hybrid Local Binary Patterns-Based Feature Extraction for Bone Age Assessment in Content Based Image Retrieval

Abstract

Skeletal maturity is evaluated visually by the comparison of hand radiographs against a standardized reference image atlas. Content-based image retrieval (CBIR) yields a robust solution without the delineation and measurement of bones. This research work comprises of five major phases: Preprocessing, texture feature extraction, relevance Score (RS) computation, bone age assessment, and similarity matching. In the preprocessing phase, noise in the image samples is eliminated by making use of Kernel Fisher Discriminant Analysis (KFDA). Texture Feature extraction is performed using Hybrid Local binary Patterns (HLBP), Epiphysis—Metaphysis Region of interest (ROI) (EMROI) feature extraction is introduced for the pre-processed image. With the aim of evaluating the bone age the Tanner-Whitehouse scheme was utilized to inspect 20 bones. For the computation of the similarity matching between query and input image samples, Discriminative Dictionary Learning (DDL) is introduced in the research work; it shows that DDL performs better when compared to other state-of the-art approaches.