Meet the team member: Chaz Hickman
By Chaz Hickman
Digital imagery is becoming more and more essential to our everyday lives, with over 500 hours of video uploaded to YouTube each minute, and over 136,000 photos uploaded to Facebook in the same time! But how can investigators begin to cope with this huge amount of content? The answer lies in image processing.
Image processing is a larger term that we use to describe many of our tools and capabilities here at CameraForensics. Meaning anything carried out on an image to gain new information, image processing may refer to a small-scale manipulation or extraction of pixels, or the complete analysis of metadata.
It enables investigative users to examine the full breadth of indexed imagery, and uncover any intelligence relevant to cases. Innovative image processing tools can help users find similar images in other online locations, uncover other images taken with the same camera, or filter by GPS location metadata to accurately pinpoint geography.
As part of our mission to transform global imaging intelligence, we’re committed to continuously enhancing our image processing capabilities. Learn what this means, and how we achieve this, below.
The best technique to use for image processing often depends on the number of images you’re facing, so using a combination of many different techniques can efficiently give us the information we need.
Our team are equipped with a unique understanding of the best techniques for processing large amounts of image data – meaning only the relevant information lands in front of investigators.
To process large amounts of images (we recently hit the 4.1 billion milestone) we’ve automated image processing techniques to take place during our initial crawl. By indexing imagery through a pipeline, we can process them using several different techniques very efficiently. Some of those are extracting the GPS data, detecting serial numbers, and identifying the camera model.
Additionally, for every image that we process, we calculate several different types of hash – unique identifiers which can be used in a variety of use cases.
This means that even if an image has no metadata attached, or has been edited, manipulated, or cropped, investigators can use them to gain valuable insights such as the other websites where similar images are being hosted.
As part of our dedication to sourcing the very best solutions, we’re always exploring innovative image processing tools. Machine Learning tools are just one example of this.
If successful, ML-based image processing tools might not simply speed up the full indexing and analysis lifecycle but introduce greater insights. Object detection and discerning new identifiers in metadata are two very real use cases – detecting unique components in each image and making them available for investigative use almost instantaneously.
However, attempting to implement Machine Learning capabilities introduces a further challenge for us:
How can we do so effectively, efficiently, and in a way that truly benefits our users? In other words, how can we avoid introducing lengthy Machine Learning processes when they aren’t needed or necessary?
Answering this involves taking a deeper reflection on the issue that we’re trying to solve.
While new ML image processing solutions hold the potential to save lives and make a profound difference, we want to make sure that we’re innovating in the most effective way possible.
To achieve this, we assess and evaluate our processes throughout development using questions such as:
As well as allowing us to develop new tools, or reinforce existing capabilities, this evaluation stage brings with it several other advantages. It gives us the perspective we need to uncover the challenges we previously hadn’t realised. It allows us to consider the question of scalability. It ensures that we always put the user first.
Ultimately, it guarantees that we truly understand and are mindful of our wider vision: to lead global online imaging intelligence and safeguard victims worldwide.
It’s always tempting to throw the latest tools and advanced analytics at an issue, but at the same time, we need to understand the intricacies and complexity at the heart of the problem. Doing this can ensure that we’re using effective practices.
Our Machine Learning research is still ongoing, and we hope to be able to share more about it with you when the time’s right. In the meantime, why not learn more about what R&D at CameraForensics involves?
With a number of tools at our disposal, we’re passionate about helping investigative teams and LEAs drive true positive change worldwide. Visit our blog for a range of insights on the latest image forensics news, trends, and developments.