Image and Video Redaction On-Premise: Our Experience

In industries like autonomous driving, mapping, smart cities, and video surveillance, where the collection of vast amounts of image and video data is commonplace, safeguarding personal information within these visuals has become crucial.

Image and video redaction, essential for processing data in accordance with major global data protection laws, has gained critical significance. While cloud-based redaction services or those provided by cloud providers might seem like the easiest options to implement, they come with valid concerns. These concerns include:

• Recent disputes between US-EU data transfer protocols
• Heightened security risks associated with transferring data over the internet to the cloud
• Bandwidth limitations when dealing with terabytes of data
• Prohibitions on transferring personal data to other jurisdictions, such as India

This article aims to share our experience of running the image and video redaction process on-premise for tens of customers, offering insights for businesses navigating the complexities of data privacy and compliance.

Security

An increasing amount of companies and municipalities are eager to enhance their data protection efforts, but they are often constrained by internal policies, or simply hesitant to adopt cloud-based or SaaS solutions for such a delicate endeavor.

In this context, we experienced different objections to the cloud:

• News about regulatory insecurity over EU-US data transfer protocols are making companies - who had previously adopted cloud services with the largest US-based providers, reconsider their usage completely or run a hybrid infrastructure with a combination of cloud and on-premise;
• Likewise, growing episodes of data breaches are making companies reconsider cloud computing because it requires data to be transferred through the internet, exposing them to possible breaches or thefts. Therefore, companies keep the most delicate data (e.g., personal data) “disconnected”, making it harder to gain access.
• Other companies have always run all their processes on-premise. Therefore, running software on-premise is a necessary condition for working with any supplier;
• For large organizations and municipalities, maintaining data locally simplifies compliance audits. In the most extreme cases, cloud-based solutions are forbidden by internal policies and regulations.

We find either one of all of these objections when talking with customers who are interested in running image and video anonymization on-premise. However, while on-premise deployment has some security benefits, it comes also with some limitations. Cloud providers invest significantly in security for their customers’ data. By running your processes locally, you need to take off this within your organization, running regular cybersecurity audits and updates.

Speed and Performance

Redaction performed on-premise eliminates the need to transfer large datasets to and from the cloud, resulting in faster processing. When internet connectivity is poor (e.g., in specific geographic locations) or simply not enough to cope with large datasets in a short period of time (e.g., when a project is time-critical), it is not feasible to move data from your cloud to the cloud infrastructure of an image/video redaction software.

To partially overcome this, it is possible to parallelize the image and video anonymization process. To better explain this, let’s define two terms:

• A "process" is a software or a task that runs on a single local machine. More powerful machines can run multiple processes (e.g., with GPUs with 16 GiB memory, it is possible to start up to 4 processes in parallel).
• An "instance" refers to a GPU.

Now, let’s consider two scenarios in which a customer uses containerized software like Celantur Container:

1. By running multiple copies of the software on a single instance, it is possible to obtain a performance boost. However, running two parallel processes may not necessarily be twice as fast, especially if the hardware reaches its capacity.

Perhaps the most effective way to enhance processing speed is by deploying multiple instances. In this scenario, speedup increases linearly with the number of instances. For example, having two instances can potentially make the process twice as fast.

2. With that being said, all of this might not be enough. In fact, the cloud has the unique benefit of scaling up the processing without purchasing and maintaining multiple local machines, data services, or workstations. Therefore, it offers better flexibility when - for instance - you have irregular data peaks (e.g., short-term projects where millions of images or video hours are collected). Also, when data volumes get just too big to be supported by local infrastructure, then the cloud is the only option.

With that being said, all of this might not be enough. In fact, the cloud has the unique benefit of scaling up the processing without purchasing and maintaining multiple local machines, data services, or workstations. Therefore, it offers better flexibility when - for instance - you have irregular data peaks (e.g., short-term projects where millions of images or video hours are collected). Also, when data volumes get just too big to be supported by local infrastructure, then the cloud is the only option.

Cost Efficiency

Last but not least, let’s talk about costs. First, eliminating the need for large-scale data transfers to the cloud can lead to significant cost savings in bandwidth and associated expenses. In addition, charges for cloud use may hide unexpected costs associated with different image or video resolutions, use of ancillary services, etc. Therefore, on-premise deployment provides a more predictable cost structure.

While on-premise deployment has some cost benefits, it also comes with some limitations:

• Managing and upgrading local infrastructure can be time-consuming and costly. Hence, you consider if you have the resources to do that;
• The processing speed is limited by the power of the local machines, which may require investments in high-performance hardware. Furthermore, for large volumes of data, scaling the processes might not be trivial.

Celantur's Solution: Balancing Security, Speed, and Convenience

Celantur Container offers a solution that strikes the perfect balance between on-premise control and flexibility. Here's how our containerized image and video redaction solution can empower your organization:

• GPU and CPU Acceleration for Efficient Processing: Celantur Container leverages the computational power of GPUs or CPUs, depending on your infrastructure, to ensure efficient and high-speed redaction. This adaptable approach allows you to tailor processing capabilities according to your specific hardware resources and cost structure.
• Seamless Cloud Integration: You can choose to run Celantur Container on your own public or private cloud infrastructure. This option provides added convenience and scalability while maintaining control over your redaction process.
• Flexible Interaction Options: Celantur Container offers flexible methods for interaction, including a REST API (widely recognized for their simplicity and ease of use among developers) and TCP connection (which gives you performance gains in specific scenarios)
• Metadata and Segmentation Masks: Unlike directly applying blur to images, Celantur Container can provide metadata (in JSON format) and segmentation masks (in JPG or PNG format) as an output.
• PGR2PGR and PGR2JPEG Conversions: As part of our commitment to streamlining your image and video redaction process, Celantur Container is optimized for PGR2PGR and PGR2JPEG conversions (native Ladybug format) through Horus' Horison Framework.