A study on the determination and depiction of the positions of the objects in the image relative to each other with the help of deep learning methods

Nowadays, with the widespread use of image processing, it is significant to automatically depict images and create subtitles. With this study, firstly Turkish-language depiction data set is intended to be introduced to the literature, and secondly, to create automatic position depiction and subtitle in Turkish language. The difference of the new data set called Konumlu-Tasvir (Positional-Depiction) from other data sets is that it contains the position information of some objects in the image relative to each other. Using the Positional-Depiction data set, deep learning methods were used for automatic depiction and captioning of images. In addition, two methods were proposed for positional depiction and subtitling, and performances were compared. For both methods, the top 10 models obtained were selected for automatic caption creation. Experimental results show that the proposed methods show successful results for automatic positional depiction and caption creation in the Turkish language. In the second part of the thesis, literature research is discussed and related studies are examined. In the third section, the definitions, models, performance measurement methods and metrics used in the study are discussed. In addition, the two proposed methods for the preparation of the new Turkish Positional-Depiction data set and automatic caption creation are discussed in detail. In the fourth section, the image outputs of the proposed methods, performance measurement values are given. Both the performance of the methods according to the performance measurement metrics and the output results were compared. Finally, in the fifth section, the analysis results and recommendations are discussed.