Image processing; convolutional neural network; object contours; features; classification;

This contribution describes the edge detection with usage of thedeep features. Our detector approaches the edge detection as an classifi-cation problem, i.e. classifying the edge and non-edge pixels. The wholeclassification process is divided into a two stages: the feature extraction andthe classification. In the feature extraction, the features are extracted fromwell known convolutional neural network VGG19 and are rescaled up tothe image resolution. In the classification stage, the features are fed into aneural network that classifies a given pixel. Our classification is pixel based,i.e. the features are extracted for each pixel separately.The classifier was trained on the dataset BSDS500 by Berkley Univer-sity. BSDS500 contains 500 images, each labeled by several subjects (variablenumber of subjects per the image). There are a two types of the labels foreach image. The segmentation labels and the object boundary labels. Wehave used latter for our classifier training. This model was contribution toEUSFLAT 2017 edge detection competition.