Edge Pixel Histograms Characterization with Neural Networks for an Improved Semantic Description

Edge Histograms are extensively used as an image descriptor for image retrieval and recognition applications. Edges represent textures and are also representative of the shapes in an image. In this work a histogram of the pixel edge directions is defined for image description. The edges detected with the Canny algorithm will be described in 4 directions. Images are divided into 16 sub-images, and a descriptor with 64 bins results. The descriptor ability for comparing images based in the Euclidean distance between histograms is going to be tested. Although the measure of the images similarity is important, it is also important to define new methods for high level description of images. The level of description can grow by defining image classes related with the image content. In this work, a neural network is used for the decision process of assigning each image to a set of defined image classes.