Bionic Eye

Experience looking through a bionic eye. This app turns a live camera feed into a low-resolution phosphene display that simulates how a visual implant might present vision. It rotates the image, extracts brightness and edges, compresses the scene into an electrode grid, and maps each cell to a brightness value that becomes the on-screen output. The main challenge is reducing a detailed camera image without losing important shapes, boundaries, and contrast. To solve that, the pipeline uses luminance extraction, edge detection, grid sampling, and spike-rate encoding before rendering the final phosphene map. The transformation process begins with a rotated camera frame that is converted into a uniform raster buffer, then passed through optical degradation, luminance extraction, center-surround filtering, and orientation-sensitive edge reinforcement. Each stage compresses the scene while preserving high-information structure: edges, discontinuities, and local contrast. The result is a retina-cortex surrogate that resembles a sparse biological encoding rather than raw pixels. From there, the image is projected onto an electrode lattice. Each electrode corresponds to a spatial region of the scene, so the grid is a many-to-one mapping from camera space into implant space. The **spike grid** is the intermediate neural representation: a 2D array of firing-rate values, one per electrode, where magnitude encodes how strongly that region should stimulate the visual system. It is not color and not display brightness directly; it is a surrogate for neural activation strength after retinal and cortical preprocessing. The **single electrode** is the atomic output unit of the model. For one cell, the pipeline takes its local spike rate, normalizes it into a bounded activation domain, modulates it with temporal pulse phase, and combines it with a per-electrode color sample from the separate color path. That yields the final phosphene element: a point whose opacity, chroma, and apparent intensity approximate what one implant contact might evoke perceptually. Technically, the display is therefore a spatiotemporal decoding layer over the spike grid. The grid carries neural state; the electrode cell converts that state into rendered light. This separation is important because it lets the simulation preserve biological fidelity in the upstream encoding while still producing a legible, interactive phosphene visualization downstream.