Automatic colorization without human intervention

This paper introduces a methodology for adding color to grayscale images in a way that is completely automatic. The proposed method of image colorization helps eliminating the need for human intervention. In this method the need for a source imagepsilas characteristics that should be suitable with the target image is eliminated. We use a database of pixels, rather than a database of images. Thus our method can be used with any application regardless of the target image characteristics. After experimenting with several images, the database was populated using a random sample of 20 images. Then, a semantically distinct grayscale image is used as input. Two methods were used to assess the results. First, we performed a survey where volunteers were asked to rate the plausibility of the colorings generated automatically for individual images. Second, Mean Square Error (MSE) was used to measure the results quantitatively. In most cases automatically-colored images were rated as mostly plausible and the MSE ranged between 9.23 and 46.27.