Cartoonify Image Using Machine Learning

Abstract: A rise in interest in the ability to quickly generate digital content through the use of automated systems has created a strong need for efficient methods to create cartoony images. A hybrid machine learning and conventional cartoonification algorithm was provided to convert real-world images into appealing cartoony types of images, while maintaining all of the required image features such as edges, contours and colour blends. Unlike the popular use of Deep Learning-based approaches, such as Convolutional Neural Networks (CNNs), Neural Style Transfer (NST) and Generative Adversarial Networks (GANs), the approach does not require the computationally intensive and complex training datasets. The cartoonification pipeline was designed in modules consisting of an image preprocessing stage involving colour space changes, converting colour images to grey images, applying Median Blur Filters to reduce noise, performing Adaptive Threshold Edge Detection, Colour Quantisation Through K-Means Clustering and creating a cartoon-like effect using Bilateral Filtering to smooth the image edges. By the use of bitwise AND, combine the structural edges of the image with the Simplified Colour Regions to create the final cartoony image. Since the study and framework were independently validated and used to create consistently high-quality cartoony images with the same level of quality, the studies also demonstrated the ability of the system to perform efficiently and effectively across all image types and resolutions. The framework for the Cartoonification Pipeline was built and can run on the Desktop Computer platform; an Android mobile application was created using Kotlin that allows users to generate and use the cartoon-style images generated from their own digital images; the application demonstrates the efficiency, scalability, and performance of the cartoonification pipeline in real time and in resource-limited environments.