Design 20: Computer Vision

Theme

When I think of computer vision broadly, I usually think of facial recognition, autonomous vehicles, medical imaging, and robotics.

In terms of consumer products, I initially think of photo editing, AR/VR games, and camera apps. However, looking at it more closely, features like text translation, QR code scanning, and digital filters appear in many consumer apps. Usually these aren’t standalone products but features built to support the main product.

Grid

Grid construction was challenging because half of the entries were at least 10 letters long, while another quarter were between 7 and 9 letters. In a 15×15 grid, a 7-letter answer occupies ~47% of a row or column, a 10-letter answer ~67%, and a 12-letter answer 80%. This made it difficult to fit the answers together while maintaining enough intersections so every answer remained connected to the rest of the grid.

Clues

For most themes, I try to divide the clues into major areas that represent fundamental parts of the subject. For this puzzle, most clues can be grouped into image recognition tasks, geometric camera models, and image processing.

Clues like Classification, Detection, and Segmentation are part of image recognition. Pinhole, Projection, and Homography are part of geometric camera models. Convolution, Canny Edge, and Hough are part of image processing.

When creating the clues, I made a conscious effort to differentiate between tasks, which describe the goals of a computer vision system, methods, which describe specific algorithms, and techniques, which describe broader approaches to solving a specific problem.

I tried to make the clues distinctive so that concepts don’t blend together. For example, Classification assigns a class label to the entire image, while Segmentation assigns a class label to each pixel. Detection locates objects within an image and assigns each a class label, while Tracking identifies and follows objects across consecutive images. Canny Edge identifies object boundaries, while Hough detects geometric shapes such as lines and curves.

Tradeoffs

I considered including answers like OCR and Gaussian Blur, but ultimately focused on concepts that I felt better represented the core areas of computer vision. I also tried to balance the answers across the field rather than overrepresenting any particular area.

Notes

The clues in this puzzle felt more unfamiliar to me than those found in my previous puzzles because I don’t have as much background knowledge in computer vision as I do in other computer science subfields.

I made an effort to use simple language to explain the concepts in the clues. However, for someone unfamiliar with the fundamentals of computer vision, this puzzle will admittedly be challenging. Hopefully, you discover a few new concepts and explore the field further on your own.