No shapes, just pixels
When you look at a photo you see a dog, a road, a face. A computer sees none of that directly, it sees a grid of pixels, and each pixel is just a number, or a small set of numbers, describing brightness and color at that exact point. A grayscale image is a single 2D grid of values from 0 to 255, where 0 is black and 255 is white. A color image adds depth: three stacked grids, one each for red, green, and blue intensity, called channels, so a 224 by 224 color image is really a block of numbers shaped 224 by 224 by 3, over 150,000 individual values.
This is why computer vision is fundamentally a math problem before it's a perception problem. Every task, recognizing a face, reading a license plate, spotting a tumor on a scan, starts from this same raw grid with zero built-in notion of 'edge' or 'object.' Everything the system understands has to be learned from patterns in those numbers.