A Hybrid CNN-LBPNET Approach for Improving Duplicate Image Detection
Abstract
The explosive growth and popularity of social media networks like Instagram, Facebook, etc., have intensely influenced the surge in visual data in recent years. Digital image-altering tools have made manipulation of images easier but their identification more difficult; this has created serious problems in multimedia forensics and content verification, as they anticipate the potential misuse of these images to provoke crowds and disseminate false information. The copy-move forgery method is often used to change or hide visual information by copying and moving parts of images within the same image. The lighting, color distribution, noise, and imaging circumstances of the
original and duplicated portions are quite similar, making it difficult to detect this type of alteration, and it is essential to verify the authenticity of an image before
adding it to these sites. This study proposes a hybrid CNN-LBPNET model that combines Local Binary Pattern (LBP) texture representation with neural feature learning to identify image-level copy-move forgeries. It makes use of RGB images to maintain structural and spatial information, and it transforms grayscale images into LBP texture maps to emphasize local intensity shifts and micro-texture variations. This integration creates a consistent input for the network, enabling it to effectively learn forensic cues based on texture and complementing spatial information. The proposed system accurately recognized copy-move forgeries at the image level when applied to the MICC-F220 dataset.
The CNN-LBPNET addresses and improves upon the constraints of utilizing merely CNNs or manually created features for detecting forged images by combining acquired spatial characteristics with explicit texture descriptors. By enhancing digital evidence verification, social media image authenticity, document integrity evaluation, and multimedia forensic investigations, the model’s robust detection abilities reduce the emergence of forged images as original.
original and duplicated portions are quite similar, making it difficult to detect this type of alteration, and it is essential to verify the authenticity of an image before
adding it to these sites. This study proposes a hybrid CNN-LBPNET model that combines Local Binary Pattern (LBP) texture representation with neural feature learning to identify image-level copy-move forgeries. It makes use of RGB images to maintain structural and spatial information, and it transforms grayscale images into LBP texture maps to emphasize local intensity shifts and micro-texture variations. This integration creates a consistent input for the network, enabling it to effectively learn forensic cues based on texture and complementing spatial information. The proposed system accurately recognized copy-move forgeries at the image level when applied to the MICC-F220 dataset.
The CNN-LBPNET addresses and improves upon the constraints of utilizing merely CNNs or manually created features for detecting forged images by combining acquired spatial characteristics with explicit texture descriptors. By enhancing digital evidence verification, social media image authenticity, document integrity evaluation, and multimedia forensic investigations, the model’s robust detection abilities reduce the emergence of forged images as original.
Keywords
CNN-LBPNET, copy-move forgery detection, local binary pattern, MICC-F220, image-level classifica- tion, digital image forensics.