MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202621094372 A) filed by Bhagvan Krishna Gupta; Dr. Durgesh Singh; and Dr. Vishal Sharma on August 04, 2026, for Passive Image Forgery Detection System And Method For Copy-Move And Image Splicing Authentication Using Multi-Domain Forensic Feature Analysis And Deep Neural Network Classification.
Inventors include Bhagvan Krishna Gupta; Dr. Durgesh Singh; and Dr. Vishal Sharma.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: A passive image forgery detection system and method identify copy-move forgery and image splicing in digital images without requiring watermarks, digital signatures, or any prior knowledge of image origin. The system operates entirely on the suspect image itself, exploiting artifacts introduced during the forgery process across multiple analytical domains. For copy-move detection, the system extracts overlapping block-level and keypoint-level features, applies dimensionality reduction, performs lexicographic sorting and approximate nearest-neighbor search to identify duplicated regions, and refines matches using geometric consistency filtering and morphological post-processing to delineate tampered zones with pixel-level precision. For splicing detection, the system analyzes inconsistencies in camera-model fingerprints derived from Photo Response Non-Uniformity (PRNU) noise patterns, lighting environment estimation from highlight and shading gradients, chromatic aberration profiles, JPEG double-compression artifacts in the Discrete Cosine Transform (DCT) coefficient domain, and Color Filter Array (CFA) interpolation residual statistics. A fusion module combines per-pixel evidence from all analytical pipelines into a unified forgery probability map. An embedded deep convolutional neural network, trained on a large-scale annotated forgery dataset, provides a second-stage semantic verification layer that re-scores candidate forgery regions using context- aware feature representations, substantially reducing false positives arising from naturally repetitive scene textures. Localization maps are output at full image resolution with per-pixel confidence scores. The system requires no internet connectivity, operates on consumer hardware, and produces a structured forensic report suitable for evidentiary presentation.
Disclaimer: Curated by HT Syndication.
