Gradient operators for feature extraction from omnidirectional panoramic images

• We propose a feature extraction framework for omnidirectional panoramic images.
• The differential feature detectors are extended for nonuniform resolution images.
• The overset grid system on the sphere called the Yin–Yang grid is used.
• We show its effectiveness through experiments with real images.

We propose an extension of the Laplacian operator and the class of differential corner detectors for omnidirectional panoramic images covering 360° views. It solves two problems of nonuniformity of spatial resolution and spherical polar coordinates singularity. For the first problem, we derive a Laplacian operator and a class of differential corner detectors that can handle spatial inhomogeneities in omnidirectional panoramic images using differential schemes. To solve the second problem, we use an overset grid system on the sphere called the Yin–Yang grid. This is realized based on the property that each of the modified Laplacian operator and corner detector is invariant with respect to coordinate transformation between the two spherical polar coordinate systems associated with the Yin–Yang grid system.