Trainable Agents Movement Strategies for Advertising Sign Visual Descriptors
Aleksei Samarin, Alexander Savelev, Aleksei Toropov, Alina Dzestelova, Valentin Andreevich Malykh, E. Mikhailova, Alexandr A. Motyko · Pattern Recognition and Image Analysis · 2022
Abstract We provide a complete description of our investigation into specialized visual descriptors application as a part of combined classifier architecture for advertising signboards photographs classification problem. We propose novel types of descriptors (pure convolutional neural networks based and based on trainable parametrized agent movement strategies) showing the state of the art results in the extraction of visual characteristics and related semantics of text fonts presented on a sign. To provide comparisons of developed approaches and its effectiveness examination, we used two datasets of commercial building facade photographs grouped by the type of presented business.