Feature descriptor optimization in medical image retrieval based on Genetic Algorithm

This paper presents an approach to represent and match images for retrieval in medical archives. A multidimensional low-level feature space including shape and texture is used to represent the image input. The large intensity variation and low contrast are main characteristics of the medical images. This presents a challenge to matching among the images, and is handled via an illumination-invariant representation. In accordance with this issue, we used several techniques based on Local Binary Pattern (LBP) such as Uniform LBP, Local Binary Count (LBC) and Complete LBC (CLBC) to extract texture features. Furthermore, one dimensional Fourier Descriptor (1-D FD) and 2-D Modified Generic Fourier Descriptor (MGFD) are used to extract shape features from medical images. Combining feature descriptors in content-based image retrieval (CBIR) systems, plays a key role due to improve the retrieval performance and reduce semantic gap between the visual features and semantics concepts. Hence, we present an approach based on Genetic Algorithm (GA) to optimize the contribution of each feature descriptors in retrieval process, and link a bridge between query concepts and low level features. The obtained results show that the proposed GA-based approach significantly improves the accuracy of content-based medical image retrieval (CBMIR) system.