}
Client: JagTag
JagTag needed a visual data code and computer vision system to form the core of their mobile marketing system.,
The JagTag code is unique because it works without an installed app and can reach the massive global market of customers without smartphones.
Reaching this large audience, however, came with several challenges: low-quality cellphone cameras, out-of-focus photos, cell carrier image compression, and more.
We designed the code to be visible even in blurry, low-resolution photos. In parallel, we designed the image recognition engine to compensate for adverse real-world lighting and focus conditions.
The fiducials are designed to stand out and reduce false-positive matches. While the data design of the code itself includes built-in error correction while still providing adequate address space for large deployments.
We devised a training set of photos that combined real-world photos and computer-generated examplars to train the image recognition algorithm. A scalable trainer ran the training algorithm across multiple machines in parallel.
We experimented with several approaches ranging from classical computer vision to modern machine learning to find an optimal image recognition algorithm. We also explored mobile and GPU computer vision techniques, in case we needed to embed the algorithm in an app or scale it up inside a data center.
The final algorithm uses a combination of multi-scale fast-sweeping, feature detection, image processing, and heuristics to locate and decode codes embedded inside input images.
The JagTag code was successfully used in marketing campaigns with major brands including Sony, HBO, Pepsi, IBM, and Sports Illustrated.
The resulting code and image recognition system was flexible enough to allow late-stage modifications, such as embedded graphics and branding.
Our code regonition algorithm handled campaigns with hundreds of thousands of scanned codes, with the ability to scale up to orders of magnitude more.
JagTag Creates One Of The Most Successful 2D Barcode Campaigns To Date, Over 100,000 Mobile Engagements - Mobile Marketing Watch