Image colorization using Scaled-YOLOv4 detector

Image Colorization is the problem of defining colors for grayscale images. Recently many research works have been conducted to propose fully-automatic colorization methods. However, many of these papers failed in colorizing images with multiple objects accurately. This might be because of dealing with the whole multi-object image as a single input. Following the efforts made in the last few years, this paper aims at studying the effect of preceding the image colorization with an object detection phase, such that the colorization will be made for each object individually as well as the full image. After the colorization of each object and the full image, they are fused together to reach a more accurate colorized image. In our work, we used a more accurate detector (Scaled-YOLOv4) than that used by the state of the art to increase the quality of the colorization results. Comparing our results to literature, it is found that using Scaled-YOLOv4 increases the Peak signal-to-noise ratio (PSNR) by 2.6%. Results of colorized images with different extensions are compared, and png extension got 5.8% better value of Learned Perceptual Image Patch Similarity (LPIPS) metric than JPEG.