Black and White Image Auto-Colorization System with RCNN

Project Summary

Development of a system that automatically colorizes black and white images using machine learning.
By using RCNN for object region identification, we identify the extent of objects
and apply higher saturation weights to areas significant to humans, achieving a high-precision colorization system.

RCNN Object Detection

RCNN divides the input image into multiple regions and extracts features from each to efficiently identify object positions and shapes. Accurately identifies important areas like people and buildings.

Saturation Weighting

Generates realistic and eye-catching color expression by increasing saturation weights for important areas like people and buildings. Maintains balance with modest saturation for backgrounds.

Lab Color Space

Uses Lab color space to infer lightness (L) from the two-dimensional information of green-red component (a) and blue-yellow component (b). Achieves more natural and accurate color expression.

State-of-the-Art Accuracy

Quality confirmed through quantitative evaluation methods such as PSNR and SSIM. Achieved the highest level of colorization accuracy as of 2021.

Technology Stack

Machine Learning Technology

RCNN (Region-based CNN)
Deep Learning
Object Region Identification
Feature Extraction

Dataset

ImageNet
Large Image Collection
Black and White Image Set
Diverse Environmental Data

Evaluation Methods

PSNR Evaluation
SSIM Evaluation
Lab Color Space Processing
Quantitative Quality Assessment

Achieving High-Precision Colorization with Cutting-Edge Machine Learning

▶︎ Object Recognition with RCNN

By adopting RCNN (Region-based Convolutional Neural Network) for object region identification, we identify important objects in images to achieve more natural and meaningful colorization. RCNN divides input images into multiple regions and extracts features from each to efficiently identify object positions and shapes.

▶︎ Intelligent Color Application

Particularly, we generate realistic and eye-catching color expression by increasing saturation weights for important areas like people and buildings. In the colorization process, we adjust hue and saturation for each type of object region to create a realistic impression. We applied high saturation to important areas and set modest saturation for backgrounds to maintain balance.

▶︎ Technical Features

Adopted RCNN for object region identification to accurately identify important objects
Used large image collections from ImageNet and their black and white versions
Improved model versatility by covering diverse environments
Improved learning efficiency through image resizing and normalization in data preprocessing
Used Lab color space to infer lightness (L) from two-dimensional information of green-red (a) and blue-yellow (b) components
Adjusted hue and saturation for each type of object region
Quality confirmation through quantitative evaluation methods such as PSNR and SSIM
Achieved the highest level of colorization accuracy as of 2021

▶︎ Application Fields

This technology can be utilized in various fields including restoration of historical black and white photographs, digital remastering of old movies, and digitization of archive materials. Applications in cultural heritage preservation and educational fields are also expected.

Looking for a System Like This?

IMAS TECH develops machine learning and AI image processing systems tailored to your business needs.

Contact Us