This paper describes a novel approach for affine invariant region detection and description. At the detection stage, a hierarchical clustering mechanism is employed to group image pixels into regions. This process is based on the Bounded Irregular Pyramid (BIP) and takes into account a colour contrast measure, internal region descriptors and attributes of their shared boundaries. High-contrasted regions are selected as salient regions. On the other hand, geometrically and photometrically normalized regions are represented by a kernel-based descriptor. The lenght descriptor is reduced by applying Principal Component Analysis (PCA). The protocol proposed by Mikolajczyk et al. has been conducted to compare the proposed approach with other similar methods. Experimental results prove that the performance of our proposal is high in terms of computational consuming and distinguished region detection and description abilities.
Salient regions; Feature detection; Affine invariant regions; Feature description