Case Study (MatchBox)
CEWE
From research to a functional MVP: solving a unique challenge for CEWE using an unconventional combination of computer vision, hardware, and software engineering
We managed to grasp a highly specific client challenge and combine several technologies in a way that goes beyond common software development practices - the result is a functional MVP ready for real photolab operations.
Client and starting point
CEWE is one of the largest European photo product manufacturers. A key part of the process is correct and fast matching of digital photos with physical outputs. This process is extremely sensitive to image quality, lighting, product variability, and differences between print technologies.
The client approached us with a challenge that had no ready-made market solution: create a CV (Computer Vision) system capable of reliably identifying real products in laboratory conditions. This task went beyond a standard software project and required a creative, research-driven, and experimental approach.
Project goals
- Validate feasibility through prototype and research.
- Find an effective combination of multiple CV methods to ensure accuracy.
- Design and build a custom capture chamber with controlled lighting.
- Integrate Canon EOS through EDSDK and build the control application on Raspberry Pi.
- Iterate the system to a stable MVP suitable for real operation.
- Create an architecture capable of growing with CEWE needs.
Our solution
The project required connecting several different technology domains into one integrated solution. Together with the CEWE development lab, we:
Step 01
Built a functional capture chamber
We designed and built a chamber with consistent lighting to ensure stable color and light conditions for capture.
Step 02
Integrated Canon EOS using Canon SDK
The solution includes a Canon EOS camera with high resolution and accurate color rendering, integrated into the capture pipeline to obtain high-quality image data.
Step 03
Developed a desktop application (Electron)
The operator desktop application in Electron enables simple capture management, communication with the chamber, and system interaction.
Step 04
Implemented advanced methods for image analysis and comparison
We included this combination in the pipeline, which enabled high accuracy even for variable products.
- OpenCV for advanced static image analysis (pre-processing, comparative statistics, and core visual characteristics).
- Embedding-based comparison using TorchVision, where we transform images into embedding vectors for much more precise similarity search and understanding of visual structure.
Step 05
Optimized matching to n*log(n) complexity
Algorithm optimization allowed the system to scale for real CEWE operational volumes.
Step 06
Designed a robust backend (Spring Boot + MariaDB)
The backend provides business logic, integration with the CEWE pipeline, and stable operation of all components.
Results
- We moved from prototype to a fully functional MVP.
- The system can accurately identify products even with high variability in input data.
- Matching runs fast enough for real photolab load.
- The solution is scalable and extensible for additional product lines.
- We successfully combined hardware, image analysis, and backend into one integrated system.
Client reference
Comming soon on Clutch
One-sentence summary
For CEWE, we created a unique system combining controlled capture, advanced image analysis, and a robust backend - and turned a complex problem with no existing solution into a functional MVP.
Need a similar delivery setup?
Book a short discovery call or send us your context. We will propose the next practical step.