Challenges in Computer Vision

In our lives, most of us have undergone X-rays and MRI scans, followed by consultations with doctors for analysis. However, these can now be easily handled by machines (AI). Secondly, we can comfortably sleep in our cars while the car itself drives for us. Next, automatically unlocking the phone using facial recognition has become a common practice. Also,if we feed an image to the machine, it can predict whether there is any object inside the image, classify the image, and also determine the position of the object in the image. Recently, using many AI applications image can be generated from a given caption only. All these things are easily possible due to the boons of Computer Vision.

Image classification is a core task of Computer Vision. However, it has many challenges:

1. Semantic Gap: For machine an image is just a big grid of numbers between [0, 255]
2. Pixel Variation: Whenever the camera moves, all the pixel value changes.
3. Illumination: Appearance of lights on images can effect greatly. Shadows, highlights, colour variations, low light can cause loss of information in images
4. Deformation: Distortions in the shape, appearance or structure of objects within images or videos can create challenges
5. Occlusion: When part of an object is hidden or obstructed by another object in a scene creates difficulties on Computer Vision.
6. Background Clutter: Presence of irrelevant or distracting visual information in the background of an image or scene can create difficulties
7. Intraclass variation: Differences among instances belonging to the same class may be a challenge

These are the challenges in Computer Vision. There is no hard coded rule for classifying an image. Hence, a Data driven approach is applied.

Steps:

1. Collect a dataset of images and labels
2. Use Machine Learning to train a classifier
3. Evaluate the classifier on new images