week0 : 1n1t14l_c0mm1t # Learning Based Image Colorization

Source: Deep Learning on Medium

Image colorization assigns colors to a gray-scale image, which is an important yet difficult image-processing task encountered in various applications. Since different colors may have the same intensity, it is difficult to solve this prob-lem using traditional methods. This paper is our proposal to the specific colorization problem.

GrayScale and Normal Image

Colorization consists of methods for adding colour to a black-and-white photograph or other image. Image colorization is still an improvable problem in the literature of image processing. Like most of the image processing problems, colorization problem has some difficulties such as illumination, different seasons, elevation angle etc. Previous solutions of this problem can be divided into two categories: scribble-based colorization and example-based colorization. Scribble-based methods use human assistance, they utilize color scribbles which are annotated by the artists on gray images. After that, their method propagates through pixels and colorizes the image. Scribble-based methods typically requires a lot of user interaction.

Gray-scale, image with colorized by Scribble-Based Colorization, Original Image

The example-based method typically transfers the color information from a similar reference image to the target gray-scale image.Example-based colorization requires identical objects for precise per-pixel registration between the reference images and the target gray-scale image. It is thus limited to objects with a rigid shape.

Deep learning based solutions got wonderful results in modeling large-scale data recently. Their learning ability can outperform humans on some computer vision and image processing problems such as classification, pedestrian detection, image super-resolution, photo adjustment etc. Proposed method directly establishes color relationships between features of the input gray-scale image and color information of the reference color image based on the corresponding training pixels. The success of deep learning techniques motivates us to explore its potential application in our context.

Original and colorizationed by Deep Neural Net.