Texture characterization via deterministic walks' direction histogram applied to a complex network-based image transformation.

•A descriptor for deterministic walks based on statistics of directions of steps is proposed.•An image transformation that describes each pixel’s relation to its neighborhood is proposed.•The proposed walk is shown to be an effective descriptor.•The proposed walk performed on the proposed image transformation is shown to improve results.•Classification rates are often better and descriptor size often smaller than in state-of-the-art methods.