How AI 'sees' objects: finding answers in biology with the haci algorithm

How do the “outline modes” in our smartphone cameras, which we use casually, or the technology that allows autonomous vehicles to recognize road boundaries actually work? For a long time, many artificial intelligence (AI) technologies have been developing around massive models trained on vast amounts of data. However, a small, efficient algorithm that closely resembles the way the human eye perceives objects has recently emerged, drawing attention. That algorithm is ‘haci’.

Why is this important?

Image processing technology is at the core of modern digital society. Recognizing objects in photos or performing optical character recognition (OCR) to read text in documents all start with finding the ‘edges’ of an image. Many existing edge detection technologies have required high computational power or often incorrectly identified background noise as edges in complex images.

On the other hand, haci uses a ‘compact’ and ‘distributed’ approach. This means that even without high-spec computers, it can recognize object boundaries lightly and quickly in environments that need real-time data processing, such as smart devices or robotic sensors. It is a foundational technology that allows the devices in our daily lives to ‘see’ their surroundings more intelligently.

Easy to understand: A conversation between pixels

The haci algorithm was designed with inspiration from the human visual system [Source 4]. To understand this, imagine the pixels, the smallest units of an image, as ‘sculptors’.

1. Comparison with neighbor pixels: In the first step, haci compares each pixel with its 8 neighboring pixels [Source 5]. This is similar to a sculptor checking the status of nearby colleagues to confirm, “Is this a boundary?” At this point, brightness (illuminance) and color (chrominance) information are used.
2. Harmonic Field: As comparison results are gathered, each pixel creates ‘harmonic field’ information as if it were trying to harmonize with its surroundings.
3. Summary by region: It collects pixel information on small grids to determine the ‘dominant orientation’ of each region, deciding in which direction it is tilted [Source 5].
4. Flank suppression: This is the most interesting part. Complex patterns or grid textures can look like unnecessary edges, but haci divides cells based on parallel outlines to suppress them. Just like removing unnecessary filters from a photo, it leaves only the true outlines of objects [Source 4].

In short, this algorithm recognizes objects by “constantly communicating with surrounding pixels, ignoring overlapping patterns, and emphasizing only the sharp boundaries.”

Current status

‘haci’ is an algorithm recently released by a developer named RenderBear, and it is a fully distributed system that extracts edges from RGB images [Source 4, Source 5]. It is currently being actively shared after being recognized for its efficiency and unique approach on GitHub and the developer community ‘Hacker News’ [Source 2, Source 9]. While it doesn’t yet have as many features as commercialized large AI models, it fully demonstrates its potential as a small, powerful algorithm optimized for a specific purpose (edge extraction).

What will happen in the future?

Research into image processing utilizing biological principles has been steadily continuing [Source 8]. If algorithms like haci become more sophisticated, it is expected that small IoT devices or drones with limited power supplies will be able to dodge objects more accurately, or that the battery consumption of augmented reality (AR) devices that recognize surroundings in real-time can be dramatically reduced. It will become a reliable assistant in the era of ‘edge computing,’ where devices judge objects themselves without sending data to massive data centers.

MindTickleBytes’ AI Reporter Perspective

AI does not necessarily require massive data centers and trillions of computational parameters. When there are more ‘small AIs’ like haci that borrow wisdom from biology to extract only the essentials, the technology around us will become more transparent and efficient. Technology that subtracts rather than adds complexity might end up defining the AI of the future.

References

1. Nuxt HN

   Show HN: My compact, bio inspired distributed edge detection algorithm (https://hn.nuxt.dev/item/50371886)

2. My compact, bio inspired distributed edge detection algorithm (https://markethunt.app/product/my-compact-bio-inspired-distributed-edge-detection-algorithm)
3. GitHub - RenderBear/haci: A compact, fully distributed … (https://github.com/RenderBear/haci)
4. GitHub - RenderBear/hci (https://github.com/RenderBear/hci)
5. hckr news - Hacker News sorted by time (https://hckrnews.com/)

6. All	Search powered by Algolia (https://hn.algolia.com/?query=Show+HN:+My+compact,+bio+inspired+distributed+edge+detection+algorithm&type=story&dateRange=all&sort=byDate&storyText=false&prefix&page=0)

7. Bio-inspired feature enhancement network for edge detection (https://link.springer.com/article/10.1007/s10489-022-03202-2)
8. Edge Detection Techniques in Image Processing - Springer (https://link.springer.com/subjects/edge-detection-techniques-in-image-processing)
9. Hacker News Summary (https://hackernews.betacat.io/)