Fence-like Quasi-periodic Texture Detection in Images

Abstract

The focus of this article is on automatic detection of fence or wire mesh (a form of quasi-periodic texture) in
images through frequency domain analysis. Textures can be broadly classified in to two general classes: quasiperiodic
and random. For example, a fence has a repetitive geometric pattern, which can be classified as a quasiperiodic
texture. Quasi-periodic textures can be easily detected in the frequency spectrum of an image as they result
in peaks in the frequency spectrum. This article explores a novel way of de-fencing viewed as a quasi-periodic texture
segmentation by filtering in frequency domain to segregate the fence from the background. A resulting de-fenced
image is followed by support vector machine classification. An interesting application of the proposed approach is
the removal of occluding structures such as fence or wire mesh in animal enclosure photography.