Image recovery of objects emitting fluorescent light in scattering media based on optical projections combined with sorting and classification processing

Imaging through scattering media is challenging due to the multiple scattering of light which severely degrades image quality and obscures hidden objects. This study experimentally validates a practical approach for imaging and classifying fluorescent objects embedded within scattering media by combining optical and computational techniques. A lens array was utilized to capture single-shot images of fluorescent objects from different viewpoints embedded between layers of biological tissue and illuminated by laser light. The resulting sub-images were extracted, digitally cropped, and evaluated using a contrast-to-noise ratio (CNR) metric. A sorting algorithm ranked the sub-images from high to low quality based on their CNR values. High-quality sub-images were aligned to a common center and averaged, excluding those with low CNR, to enhance image reconstruction. A support vector machine was trained on reference images to facilitate subsequent classification during the reconstruction process. High classification accuracy was achieved for fluorescent objects of varying geometric shapes.