Blurring Faces in Videos Without Paid Tools: A Simple Python Solution for NGOs

An NGO faced challenges in protecting data privacy by blurring faces in a video. After finding existing tools inadequate, they developed a DIY solution using Python and OpenCV to address the issue without relying on paid software. The organization needed to anonymize specific individuals in a video to safeguard sensitive stories, but traditional editors charged high fees, and online tools lacked precision. Many platforms only allowed full-frame blurring, which wasn’t suitable for selective masking. Others imposed limits on export credits, complicating workflows. To tackle this, the team devised a manual process rooted in video fundamentals. A video is essentially a sequence of images, typically 24–30 per second, similar to a flipbook. By splitting the video into individual frames, they could edit specific elements before reassembling the footage. The approach involved using OpenCV to extract frames and Ikomia for face detection and blurring. A Python script processed each frame, applying a blur to detected faces. Blurred and original frames were saved separately, allowing manual review to refine results. For instance, problematic frames could be replaced with the previous one to maintain audio synchronization, ensuring the final video remained intact. While the method offered flexibility, it had limitations. It wasn’t ideal for long videos or large-scale projects, where professional tools might be more efficient. Additionally, it couldn’t handle multiple faces in a single frame without adjustments. However, the solution could be enhanced by integrating face_recognition libraries to distinguish between known and unknown individuals, enabling targeted blurring. Manual oversight proved critical. Though automation streamlined the process, human review ensured ethical decisions—such as blurring not just faces but also sensitive text or objects in the background. This step added trust and adaptability, allowing the NGO to tailor the anonymization to its specific needs. The entire workflow ran in Google Colab, eliminating the need for software installation. A notebook provided the code, making the process accessible for free. While not a perfect fix, the approach demonstrated how creative problem-solving can address real-world challenges, especially for organizations with limited resources. For NGOs or others facing similar issues, this method highlights the value of combining open-source tools with careful human intervention. It balances efficiency, cost, and ethical considerations, offering a practical alternative to commercial solutions. The process remains a work in progress, but its simplicity and adaptability make it a compelling option for those prioritizing privacy without compromising quality.