Multimodal and Mirror Invariant image description

This paper presents a novel local interest region description method named Multimodal and Mirror Invariant Local Pattern (MMILP) which is specially designed to be invariant to mirror and modality image variations. However, the proposed descriptor is also invariant to other image transformations such as rotation, scale, illumination. MMILP also avoids the dominant orientation calculation which is an intermediate step to normalize the rotation effect. Therefore, the descriptor building time is comparatively lower than the other local feature descriptors. The experimental results prove the efficiency and effectiveness of the proposed method. MMILP can also be used to classify the multimodal and mirror variant textures.