Large-scale camera/image search engine
Nowadays, billions of pictures and videos are acquired everyday due to the pervasiveness of smartphones, tablets and digital cameras. This huge amount of visual information is dramatically modifying the way forensic investigations are performed. Photos and videos represent a large body of available information, which can be used as evidence connected to a crime. It is often the case that during investigations it is necessary to trace the origin of a photo to the device that acquired it. As an example, this allows to assign responsibility for images with illegal content, or find connections among people suspected of a crime.
Current technologies allow to extract from a given picture the information related to the camera that shot it, in a such a way as to produce reliable forensic evidence. Unfortunately, state-of-the-art methods do not scale efficiently to the vast amounts of images that are routinely handled during an investigation. This limitation prevents certain types of analyses, such as those requiring cross-linking between a database of cameras and a collection of millions of pictures.
ToothPic Search provides an effective tool for law-enforcement agencies during investigations and it is the first forensic instrument that allows to link a device to all the pictures acquired by it in a large-scale collection. The innovative patent-pending technology developed by ToothPic makes camera identification possible on unprecedented scales. Currently, no equivalent technology exists on the market.
Camera identification refers to the ability to link a picture to the device that shot it. This is currently possible by means of two techniques:
Textual metadata (e.g., EXIF headers), the most common technique, do not represent a reliable forensic evidence, since they are easily tampered with or they may be missing e.g. when a picture is shared through a social network.
Photo Response Non Uniformity (PRNU). PRNU is an invisible pattern due to manufacturing imperfections of camera sensors which constitutes a unique fingerprint of the camera and can be extracted from photos to reliably identify the specific sensor that acquired that photo.
PRNU fingerprint analysis is the preferred technique during investigations since it is a robust and reliable evidence. However, state-of-the-art techniques for this kind of analysis suffer from large complexity since the noise-like characteristics of PRNU patterns do not allow efficient processing with known techniques.
ToothPic Search is enabled by a proprietary PRNU fingerprint compression technique and related hierarchical retrieval technology allowing to scale the system so to handle image databases of unprecedented size. The technology is not only efficient, but also robust to image rotation, rescaling and the most common automatic crops.
The software allows to:
- Search by device: this search mode will return all the pictures in the collection acquired by the query device, along with a reliability score for each picture.
- Search by photo: this search mode will return the device that has shot the query picture, if present in the database, along with a reliability score.
ToothPic Search is a camera identification product that does not rely on textual metadata associated to pictures, as those are unreliable as forensic evidence or often missing. Instead, it is based on the optical sensor PRNU fingerprint, which is a physical untamperable characteristic. Moreover, it represents a reliable tool for forensic investigations since it is resilient to common image transformations applied by photo sharing software, instant messengers and social networks, such as rescaling, rotation and automatic image cropping.
The proprietary compression and search technology allows ToothPic Search to be the first camera identification software able to scale to databases of millions of pictures and devices. Moreover, it is based on a distributed backend architecture whose performance scales linearly with the number of workstations employed.
Intelligence agencies as well as forensic investigators can use this tool to build an archive of devices and pictures collected or seized from various sources. In this sense, the collection of pictures and devices grows in time and may allow to detect non obvious connections among cases. This is an innovative feature enabled by the large-scale capabilities of ToothPic Search. As the database size grows, due to accumulation of digital evidence from past cases, it is possible to dynamically expand the architecture with additional workstations.
The client-server software architecture allows to build a large collection of pictures and devices which is shared and concurrently accessible by multiple clients. This is in contrast to existing products for camera identification, which were intended for small scales, thus adopting stand-alone packages to be installed and used on a single machine.