AI in image processing and face recognition? It’s getting hot in here!

There’s a lot of real estate in our brain dedicated to facial recognition and to physics that takes a lot of computational power of our brain. — Jon Favreau

Perception through vision has constantly played a crucial role in human life, and images continue to be one of the most important information carriers despite rapidly varying technological environments. Artificial Intelligence, at this point, is much like a toddler. Computer vision gives it some sense of sight, but that doesn’t come with an inherent understanding of the physical universe. Driver-less cars, for instance, use computer vision and image recognition to identify pedestrians, signs, and other vehicles. So let’s talk about the major aspects that potentially drive AI to be more relevant to the real-world: Image processing and recognition.

Image recognition is a hot and hyped topic in machine learning, artificial intelligence, and other technology circles. Amidst the hype, organizations large and small are trying to understand the specific industry use cases potential for the variety of underlying techniques and processes involved in image recognition.

Readers often tend to mistake image processing for image recognition. So let’s learn the terms first!

Image processing refers to a two-step process for preparing the image files for analysis and the subsequent fitting of algorithms to make selections or assign perceived qualities. Now here lies the difference. The steps taken in image processing will directly impact whether downstream recognition will be feasible.

Face ID technology performs user authentication and can be extended to many use cases viz., unlocking devices like smart devices and systems, authorization and validation of online transactions, mass surveillance in heavily crowded public areas. In fact, it turned out to be better than traditional facial recognition and a much more effective tracking tool than fingerprints since it doesn’t require physical contact for identification.

Sounds familiar? Of course it should, these are systems that we use in our daily and hence we need to understand them.

Image recognition, due to its complexity, usually requires a compound model. On reading a bit on CNN, it begins to make sense that these algorithms have been most successful in image recognition challenges. In these highly complex neural frames, there is a certain number of overlap between receptive fields of individual neurons, similar to the human visual cortex. Let's learn more on CNN…

Convolutional Neural Networks: A technique that overcomes large buffers & computational processes, has mutable manual parameters and is efficient in terms of space complexity due to the usage of graphics accelerators. It encompasses the following methods…

Insight on the working of CNN Model

1. Residual Learning: Each subsequent layer in the NN is responsible for, in effect, fine-tuning the output from a previous layer by adding a learned “residue” to the input
2. Batch Normalization: Technique for training CNN that standardizes the inputs for each layer, so they learn on a more stable distribution of inputs, thus accelerates the training of the network
3. Rectified Linear Units (ReLU) : Activation function : y = max(0,x)

How does image recognition work?

Let’s consider face recognition, in particular, to understand the concept better and break this down step by step.

1. Detection. Trace the user’s face in the image and delimit it with a bounding box.
2. Alignment. Normalize the face with the database, such as geometry and photometry.
3. Feature Extraction. Extract features from the face that can be used for recognition

Now we are ready to recognize the face and voila!

Although it is well-known that these systems have countless uses and unparalleled relevancy in real-life scenarios, there still are a few unanswered questions…

• Don’t these systems invade privacy?
• How secure is all this personal data?
• Racial Bias in FR systems?! How bad can this sound?
• Is it completely reliable?

While researchers are working on that…let’s not forget how facial recognition software can pick a person out of the crowd, but the vending machine at work can’t recognize a dollar bill with a bent corner! So yeah, we’ll work on it too XD

If these insightful individuals are predicting doomsday with AI, should we be taking this much more seriously?

“The development of full artificial intelligence could spell the end of the human race.” — Stephen Hawking

“AI could create an immortal dictator from which we can never escape.” — Elon Musk

Fortunately, researchers are already attempting to solve that problem. Defense Advanced Research Projects Agency (DARPA), is overseeing the aptly named Explainable Artificial Intelligence program. The goal is to enable human users to understand, appropriately trust, and effectively manage the emerging generation of artificially intelligent partners.