Method for Speckle Noise Reduction

Unmet Need
Although laser light is generally cleaner, easier to collimate, and more efficient for use in imaging than incoherent sources (LED’s, lamps) it is rarely used. This is in large part due to the presence of speckle noise pollution inherent in laser-illuminated images. In the past, reduction of this noise post facto was impossible. The only robust method for preventing the presence of dense speckle noise were extremely expensive and cumbersome optical laser speckle noise reducers. Consequently, most optical imaging systems continue utilizing LED and Arc lamps for illumination, despite them being less efficient, larger and difficult to collimate. This the case in numerous applications, including medical endoscopy, industrial borescopes, machine vision systems, photography, and laboratory research.  
 
Technology Overview
Utilizing deep neural networks, researchers at Johns Hopkins have established methodologies to digitally strip images of dense laser speckle noise after imaging. During a brief training period neural networks are developed and trained to recognize the difference between laser-illuminated images and noise free LED-illuminated images. Then, using statistical transformation the technology is capable of recognizing, and subsequently removing noise present on laser-illuminated images. Initial testing shows results on par or better than optical laser speckle noise reducers, for a fraction of the cost and without the need for new cumbersome hardware. This technology positions laser illumination as a cheaper, easily collimated, and functionally superior alternative to modern industry standards.
 
Stage of Development
Functional prototype, extensive testing

Category(s):