Object detection in images of natural scenes represented by AR models using Laplacian pyramids: application to leather defects localization

A methodology for object detection and localization by Laplacian pyramid analysis of the features of AR (autoregressive) models applied to 2-D iconic images of natural surfaces is described. A symbolic image of a leather defect was built with features of simultaneous autoregressive models. Laplacian pyramids were then implemented for detecting defects of calf leather patches, on different resolution levels. Strategies for enhancing the wrinkled patches of the leather are discussed based on the parameters of the models. Thresholding the Laplacian pyramids for noise filtering is studied taking into account the histograms of each Laplacian image. Probable defective patches were marked by squares on a simulated original iconic image