It's all a blur
9 hours ago
- Blurring an image by averaging pixel values is not necessarily irreversible; information can leak due to window boundaries and discrete stepover.
- A one-dimensional moving average blur can be reversed by solving equations that relate blurred pixels and known padding values, enabling reconstruction of original pixels.
- A right-aligned moving average window simplifies the reconstruction process, allowing iterative recovery of all pixels without gaps.
- Two-dimensional blur (e.g., box blur) is harder to reverse but can be improved by biasing the current pixel in the average, enabling recovery even from lossy JPEG images.
- For constrained data like text, pixel-level reconstruction may not be needed; comparing blurred regions to known symbols can suffice for identification.
- The success of reconstruction depends on knowledge of the blur algorithm, boundary conditions, and the presence of quantization noise.