Paper Title
Hybrid CNN-OCR System for Mixed Content Image Classification

Abstract
With the rapid rise in image sharing through social media and digital communication platforms, as shown in fig 1.0, there's an increasing need for smarter systems that can effectively categorize and filter image content. A large portion of these images falls into two key categories: real captured photos and text-based graphics. Real images typically include photographs of people, environments, or objects and often carry emotional, cultural, or informative value. In contrast, text-based images—like quotes, greetings, promotional content, and announcements—primarily consist of written messages, sometimes enhanced with decorative graphics. While these text-heavy visuals can be engaging, they often clutter digital photo libraries and take up unnecessary storage space. To address this issue, this project presents a hybrid classification approach that can automatically sort and filter both real and text-based images. The system combines the strengths of Convolutional Neural Networks (CNNs) and Optical Character Recognition (OCR) to boost classification accuracy. The CNN is trained to distinguish visual features that separate real-life photos from text-centric content, while the OCR module provides deeper insight by detecting text presence, density, and meaningful keywords. By integrating these two technologies, the system delivers a more reliable and scalable method for image classification. This dual-layered design not only improves digital content organization but also helps optimize storage by reducing clutter. The proposed framework is adaptable for use across different platforms and devices, offering users a more streamlined and organized digital experience. Moreover, by integrating visual and semantic analysis, the system can be adapted to various applications, including digital archiving, photo gallery optimization, and intelligent content moderation. Its modular design allows for extension into multilingual OCR support and cloud-based classification services, making it a forward-looking solution for the evolving landscape of digital media management. The hybrid CNN-OCR architecture thus bridges the gap between visual computing and text recognition, providing an intelligent and context-aware framework for handling heterogeneous image datasets at scale. Keywords - Image Classification of real and text-based images, Hybrid OCR-CNN, OCR, CNN, Image filtration