Histology Image Classification, Retrieval Using Extracted Local Features From Circularly Surrounded Neighbors

With the rapid advancement of digital imaging technologies, histology image analysis has become increasingly important in medical research and clinical diagnosis. Histology images are widely used for both educational purposes and disease detection, particularly in oncology. Computer-aided diagnosis systems offer valuable support to pathologists by automating image analysis and retrieving visually similar cases for comparison. In this study, a content-based image retrieval and classification framework is proposed using handcrafted texture features extracted from histopathological images. The method incorporates rotation-invariant uniform Local Binary Pattern (LBP), first and second statistical moments, and Local Variation Code (LVC) features computed over circular neighborhoods with multiple radii. These features are designed to capture local micro-patterns while being invariant to rotation and grayscale variations. The performance of the proposed approach was evaluated using two publicly available histopathology datasets: Kimia Path24 and BreakHis. Experimental results showed that even individual feature types can achieve high classification accuracy on one of the datasets. Moreover, combining different feature sets further improved performance, especially in modeling fine-grained structural differences. To enhance discriminative power and reduce redundancy, several feature selection techniques were applied. The resulting low-dimensional feature representation not only improves computational efficiency but also outperforms previously reported methods in terms of retrieval and classification accuracy on the tested datasets.