Novel Computer Vision Approach for Scale-Specific Generative Stick Figure as Synthetic Tribal Art Works

Abstract

A detailed methodology of generation of figures of Warli art (an ancient tribal art) using precise computational geometry, advanced trigonometry and intelligent computer vision techniques, has been presented in this chapter. The figures have been generated for both the genders- male and female. The visual variances between the genders are attributed to carefully chosen phenotypic traits and differences in anatomical structures, that have been incorporated during generation of the images. A total of 60,000 images have been generated in multiple batches, with both genders in almost equal proportions. The batches of images generated are properly labelled and annotated and finally split into training and validation data. The whole directory structure has been carefully formatted so that it can be used effectively as data set for future model training. In-fact, the data set generated has been used to train Generative Adversarial Networks (GAN) based models and has produced promising results. The entire work is done using spherical polar coordinates, instead of complex fractal geometry. All the sequential steps of the closed loop operation have been performed to generate aesthetically pleasing figures, viable for multi facet uses in the future.