InXeption, developed by HTX’s Sense-making and Surveillance Centre of Expertise, makes blurry images a thing of the past for Home Team officers.
There are countless memes on social media sites mocking the pixelated quality of CCTV footage and how supposedly “expensive” CCTV cameras often provide footage that rivals the photo quality of mobile phones from a decade ago.
However, blurry CCTV footage becomes significantly less funny when you realise that it could hinder investigations bringing criminals to justice. After all, identifying a criminal when their face looks like a pixelated mess is challenging. While commercial software can be used to enhance existing images or footage, this process is often time-consuming.
At the Home Team, every second counts in solving crimes. Clear evidence is required to swiftly apprehend criminals as it reduces the time needed for officers to verify key details.
Unfortunately, the quality of certain images and videos are not within the Home Team’s control. For example, if a robbery occurs at a bakery, the CCTV footage used to apprehend the criminal would be from the bakery’s CCTV cameras and could be unclear. This is where InXeption steps in to swiftly enhance photos or videos by increasing their brightness, reducing motion blur, and increasing the resolution of faces.
Complexities surrounding image enhancement
Prior to InXeption, Home Team officers used commercial software to enhance the quality of evidence. Such commercial software is not as easy to use as Instagram’s filters, where one can simply move around the various sliders until the image looks decent enough. Instead, officers had to meticulously follow many precise steps to improve the quality of a single image.
As you can imagine, enhancing a single image was time-consuming to begin with. However, Home Team officers often had to replicate the process for hundreds of other unclear images, detracting from their other crime-solving efforts.
Simplicity is key
Where InXeption shines is through its simplicity. While commercial software required Home Team officers to follow numerous complicated steps, InXeption gets the job done in less than three steps. Home Team officers simply need to upload the images they wish to enhance, select the type of enhancement (i.e. auto enhancement, low-light enhancement, motion deblur or facial super resolution) and run the program!
Auto-enhancement is the most time-saving feature, as Home Team officers do not even need to identify the quality issues with the images and choose the appropriate types of enhancement. Under auto-enhancement, InXeption’s deep-learning algorithms can identify images exhibiting three types of quality issues—namely, low light, blurry photos, and low-resolution photos. Once these quality issues are picked up, InXeption will then proceed to automatically apply the respective quality enhancements to the photos.
InXeption's low-light enhancement feature lights up the crowd. (Photo: HTX)
For example, CCTV footage may have documented a theft occurring in the dead of night. Naturally, this would cause the footage and stills from the footage to be extremely dim. Once InXeption picks up on the lack of light in the image, it uses low-light enhancement to “light up” the image and make it clearer.
InXeption's motion de-blur technology brings clarity to blurry features, such as license plates. (Photo: HTX)
Of course, InXeption does not just work on CCTV footage. The Singapore Police Force (SPF) often receives a treasure trove of evidence from citizens’ reports through public channels. However, some of these key pieces of evidence may have been taken in the heart of action, resulting in the “motion-blur” effect. For example, if a bystander whips out his/her phone to capture the license plate of a vehicle involved in a hit-and-run accident, the image will naturally have a “motion-blur” effect due to the vehicle’s speed. To increase the clarity of such images, InXeption uses “motion de-blur” technology to remove the blurry effect from moving objects.
Say goodbye to pixelated faces with InXeption's facial super resolution technology. (The original image (left) is taken from “Progressive Growing of GANs for Improved Quality, Stability, and Variation – Official TensorFlow implementation of the ICLR 2018 paper” by Tero Karras (NVIDIA), Timo Aila (NVIDIA), Samuli Laine (NVIDIA), Jaako Lehtinen (NVIDIA and Aalto University) and is licensed under the Creative Commons BY-NC 4.0 licence. The enhanced image (right) by HTX is a derivative of the original image and is licenced under Creative Commons BY-NC 4.0 licence.)
One of the biggest gripes people have against CCTV footage is that faces are captured at an extremely low resolution, sometimes to the point of looking like Minecraft characters. Here, InXeption increases the clarity of pixelated faces through facial super resolution technology, rectifying pixelation to reveal facial details.
InXeption was recently showcased at the Milipol Asia-Pacific – TechX Summit exhibition held at the Marina Bay Sands convention centre from 3-5 April 2024.