The human retina provides localized convolution (Mexican-hat) gradient detection; machine vision uses the same methods. Once local gradients are detected, the retina uses its parallelism (100Million pixels) to examine local, medium and large scale "textures"; upsets to the "texture" become very interesting to humans. Shape-from-texture and shape-from-monocular-occlusion lets humans derive a 3_D model of the world.
Machine vision does not yet have this 100Million-pixel-in-parallel processing.
Regarding the process of sensor-to-pixel-on-screen, the sensor uses a grid of diodes, these diodes having sufficient junction depth to with some likelihood absorb the photons and collect an electron or two or thousands. DarkCurrent sets the detection floor. Periodically, these diodes are flushed of their electrons as the electrons are converted into voltages and those voltages become quantized numbers, sent offchip to various compute engines. Or your screen.