We humans see objects, places and people with our eyes. We are gifted with a natural tool for object analysis, detection that helps us identify things nearby. But have you ever wondered how Face Lock works on both Android and iPhone? Do they also have computers that look around the world like humans?
Computer vision
Computer vision is a type of artificial intelligence that helps a computer see the world, interpret and analyze the visual world. It also uses the concept of machine learning to recognize the different objects it sees and classify them with similar objects. The machine learning model used here is already well trained to do this job.
But, in the process of identifying and classifying objects, several difficulties can have a great effect on the final result.
1) Loss of information during 3D to 2D conversion
In this case, when the object is captured by the camera, the main problem is the pinhole we use. A pinhole is a box with a small hole used for perspective projection.
The real trouble with the pinhole model is that when the image is taken, the projective transform sees a relatively small object close to the camera. In this case, we humans need a ‘scale’ to predict the actual size of an object. But this won’t work for PCs.
The actual image of the object is not captured in the computer so the size of the coin, bat and building is the same when viewed as the image in the computer.
2) Interpretation
When we humans try to analyze or understand an image, we use all of our previous long accumulated knowledge and experience to fully interpret the image and gain insight from it. We have invested several years in training artificial intelligence models to understand observations, but the model’s ability to understand observations is still limited. In order to increase the level of interpretation, several mathematical tools are used.
3) Noise
Noise is present in every image measurement. We use mathematical tools that deal with such unreliability. Noise cannot be removed to some extent, but using such tools can complicate image analysis.
4) Big data
The image and audio files we use take up huge amounts of memory. A sheet of A4 paper is scanned in monochrome at 300 dots per inch, which corresponds to 8.5 MB. Non-interlaced RGB 24-bit video in color 512 * 768 pixels, making a data flow of 225 MB per second.
If the processing we do is not very simple, then it is difficult to achieve real-time performance such as processing 25 to 30 frames per second.
5) Local view versus global view
An image analysis algorithm analyzes a small storage in local memory like a pixel in an image, the computer sees the image through a keyhole. When we look at the picture through the keyhole, it is more difficult to understand what the picture shows. But it is easy for people to interpret an image if it is seen globally
Conclusion
In this blog, you can get a clear picture of the various difficulties you face while processing images using computer vision. Once we overcome these difficulties, we can make computer vision accessible to everyone.
I hope you enjoyed reading this blog. Like and comment your views on today’s topic. Go to my profile for more such blogs.
Happy studying!!