Computational imaging · 2026 · Solo — signal processing, API, interface, deployment

ImageManip

A browser and API image processor that composes signal-processing stages into one inspectable production pipeline.

The problem

Image transformations are often presented as one opaque filter, which makes it difficult to understand how frequency content, sensor characteristics, compression history, entropy, and metadata interact. ImageManip exposes those stages as one bounded browser workflow and a documented JSON API. It is an experimental transformation tool, not a guarantee of detector evasion or image provenance.

Architecture

Key decisions

01

Compose frequency-domain and pixel-domain stages

FFT spectrum changes handle spatial-frequency structure, while lens falloff, perturbation, sensor noise, and entropy conditioning operate in pixel space. Keeping them as explicit stages makes the transformation pipeline inspectable.

02

Expose camera profiles as product controls

Canon, Sony, Nikon, and iPhone profiles vary noise, color, pattern, lens, and metadata characteristics instead of hiding every parameter behind one generic preset.

03

Ship one engine through UI and API surfaces

The browser supports direct drag-and-drop use, while the documented JSON endpoint lets cURL, Python, and JavaScript clients invoke the same production processor.

04

Keep the claim narrower than the implementation goal

The portfolio describes the verified transformation pipeline and working API. It does not claim universal detector evasion, authenticity, or provenance because no detector benchmark is published with the project.

Metrics

4
camera profiles
3
processing intensities
2048 px
maximum input dimension
2
production API endpoints