Image Recognition: Detection of nearly duplicate images

Image recognition is used in many applications to detect images with same or different image content. This paper proposes image similarity measures. Usual method of serving photographs as evidence has been carried out for over a century. As there is digital information revolution, these methods are required to improve with same pace of time. Digital images are used and recognized in law enforcement as important tools in criminal investigations. The technique that is used in this paper is designed to identify if the image found on crime scenes is present in criminal database handled by forensic departments. Two important features of this thesis include recovering deleted images from any disk and detection of nearly duplicate images. Technique used for detection of nearly duplicate images is called as image fingerprinting, also known as image hashing. During criminal investigations, fingerprint evidence plays an important role. Image fingerprinting is defined as its literal meaning. As one’s fingerprints are unique and they represent particular human being, similarly images have unique image fingerprints. Image fingerprints can be used to identify particular image in case of crime. Perceptual Hash gives the result to find image fingerprints. The techniques used in this paper uses difference hash for given images. Hamming distance is used to check the images which are almost similar with slight modification. Fingerprints of images don’t change if an image is re-sized, compressed or expanded. Difference hash is used to identify such images. But if image is cropped or if image is taken from different angle then hamming distance is required. The paper also talks about different tools required to recover deleted images.