Computation is playing an increasingly important role in imaging and sensing. Computational imaging now comes in many shapes and sizes - at one end of the spectrum are classic algorithms such as phase-shifting and phase unwrapping that retrieve a phase image from multiple images recorded by interferometers; and at the other end are modern compressive sensing algorithms that generate an image from measurements taken by a single-pixel camera over time.

Mahalati et al., propose a novel computational imaging approach of enhancing the resolution of endoscopic imaging systems based on the mutli-mode fiber. The key ingredient of their technique is illumination of the specimen with a set of random patterns. The images produced by the specimen in response to the illumination patterns are used (in conjunction with precalibrated coefficients) to reconstruct the specimen's reflectivity.

One is naturally led to wonder why does the illumination with random patterns lead to resolution enhancement as compared to scanning of a spot across the field of view of the fiber? The authors derive the analytical point spread functions for both (pattern-illumination and spot-illumination) designs and argue that the combination of pattern-illumination and reconstruction is equivalent to spot-illumination and deconvolution - but more immune to noise.

I find a particularly interesting analogy among the structured illumination methods (SIM) used in biological fluorescence microscopy. SIM extends the lateral resolution of the imaging system by illuminating the specimen with a sequence of sinusoidal patterns - each shifted by 1/4th of the period. The sinusoidal patterns at the specimen are generated by interference of two beams travelling at high angle that meet at the specimen plane. Illumination with periodic patterns, systematically aliases the fine details in the specimen into not-so-fine details that the microscope can resolve. The fine details can be teased out from multiple aliased images. This approach of using systematic patterns to illuminate the specimen boosts the lateral resolution of imaging system by a factor of 2 - the same as achieved by present authors using random patterns.

How do these ideas of extending the resolution by factor of two compare? How effective is the illumination with random patterns for biological microscopy? Or conversely, how effective is the periodic illumination in endoscopic imaging to extend? These are some more intriguing and likely impactful questions in the realm of computational imaging.

---

---

 Add Comment

You must log in to add comments.