Method for Generating Vector Image Features for Contextual Search Tasks

This article discusses a method for generating a vector of image features for the task of contextual search. The vector consists of numerical values obtained from texture and color features, key points, values obtained from a convolutional neural network, and features of the object in the image. The criterion for image similarity is also described. As part of this work, a convolutional neural network model was developed to highlight significant visual characteristics, and the dependence of color space and visual search accuracy was analyzed. The conducted studies showed the effectiveness of the proposed method for generating an image feature vector. However, this method is inferior in terms of vector formation time, since it consists of a combination of several methods. This indicator can be improved through the use of parallel computing technologies.