A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
The visual cortex is organized hierarchically from primary visual cortex (V1) through intermediate areas (V2, V4, MT) to higher-order regions in the dorsal and ventral streams. Each stage extracts progressively more complex features—simple orientations and spatial frequencies in V1, edges and textures in V2, object parts in V4, complete objects and scenes in higher areas. This hierarchical organization enables efficient computation of visual information.
You know from the visual system and sensory cortical streams prerequisites that visual information travels from the retina through the lateral geniculate nucleus to primary visual cortex, and that cortical processing splits into dorsal ("where/how") and ventral ("what") pathways. Now let's open the hood on how each stage transforms raw visual input into the recognizable objects and scenes you experience.
V1 — primary visual cortex in the occipital lobe — is the first cortical processing station. Its neurons respond selectively to very specific low-level features: a bar of light at a particular orientation (say, 45 degrees), at a particular spatial frequency (fine versus coarse detail), in a particular location of the visual field. The receptive fields are small — each V1 neuron "sees" only a tiny patch of the visual field. This was established by Hubel and Wiesel's Nobel Prize-winning work: they discovered simple cells (responding to a bar at a specific orientation and location) and complex cells (same orientation preference but tolerant to position shifts). Critically, V1 knows nothing about objects, faces, or meaning. It is simply detecting oriented edges and luminance gradients across the visual field.
As you ascend the hierarchy — V1 → V2 → V4 → inferior temporal (IT) cortex — several things change in a systematic pattern. Receptive fields become progressively larger (neurons respond to stimulation across larger portions of the visual field). Tuning becomes progressively more complex (V4 neurons respond to colors and intermediate shapes; IT neurons respond to complete objects, faces, and scenes). And representations become progressively more invariant — resistant to changes in position, size, and lighting that would disrupt lower-level detectors. A face-selective cell in the fusiform face area fires to a face regardless of whether it's large or small, centered or peripheral, brightly lit or shadowed. This invariant object recognition is computationally remarkable — it's something that took decades for computer vision to approximate.
The hierarchical principle is elegant in its logic. Rather than having a separate detector for every possible object at every possible position, scale, and lighting condition (an impossible combinatorial explosion), the brain builds complex representations by composing simpler ones. V1 edges → V2 contours → V4 object parts → IT complete objects. This composition allows recognition of an unlimited variety of novel objects from combinations of previously learned primitives. Deep neural networks (convolutional neural networks) that achieve human-level object recognition were directly inspired by this biological hierarchy — and the learned representations in artificial networks closely parallel what is found in V1 through IT cortex, validating the computational logic of the hierarchical architecture.