Through the development of social media, we need technology to search for images created at a rapid rate. Generally searching for images can be done by users inserting tag information into the images. This study suggests a system that is to insert tag information automatically by using trained data when new images enter. A suggested system, region of interest is selected in the images and it is divided into many blocks. Histogram is made by computing Local Binary Pattern for each blocks. computed histograms are connected with each other, which is used as feature vectors, and it is trained at random forests. Each feature vectors having strong on rotation, are created by computed each histogram don by Discrete Fourier Transforms. In the same way, feature vectors are created when new images enter; then, by using Random Forest, images can be expected which categories they are involved in. Tags for certain categories are collected from tag pool and tags are automatically inserted by calculation of weighting. Performance is evaluated after a comparative experimental study between a proposed system and an existing tagging system.