No abstract

Auto-focusing is an important problem for various imaging systems. A good focus measure, or evaluation function, is critical for the accomplishment of this task. Some evaluation functions are introduced. Analysis is done on evaluation functions using global data such as the squared image gradient magnitude of the whole image, asserting that these approaches may suffer from the indiscriminate use of the data. A new evaluation function is proposed. By partitioning the image into small blocks and using the maximal squared gradient magnitude in each block to construct the evaluation value, impacts of strong off-boundary gradients deteriorating the performance of the "global" approaches can be suppressed. Experiments on a real-world microscopy video data set show that our method exhibits higher focusing accuracy.

An optical coherence tomography system is proposed for synchronized zoom imaging of the cornea, retina, and the whole eye. The system was combined with an electrically tunable lens provided with 15 ms zoom response time and a customized optical delay line. A full-range technique was used to extend the depth of the B-scan cross sectional image. The anterior and posterior segments of the human eye were scanned by a coaxial rotating double galvanometer system. The transverse scanning ranges can reach up to 8 mm in whole eye scanning and 14 mm in fast single-frame scanning. The speed of image acquisition is over 4 Hz, and five B-scans were stitched to obtain a whole eye image. The system with electrically tunable lens and optical delay line achieved whole eye depth imaging in vivo.

The retinal nerve fiber layer (RNFL) evaluation is becoming a very effective method for the clinical diagnosis of early glaucoma. The purpose of this paper is to extract the pulsations of the RNFL, which might be used as a novel biomarker for glaucoma diagnosis. To demonstrate that the optical coherence tomography (OCT) could extract the subtle RNFL dynamic pulsatile motion in normal eyes in vivo, the subjects’ retina was imaged by spectral domain optical coherence tomography (SD-OCT) based on histogram RNFL pulse extraction algorithm. Firstly, B-scan images of multiple retinal layers in normal subjects were acquired. The RNFL was identified from each B-scan with a segmentation algorithm based on shortest path and convolutional neural network. Secondly, a histogram-based RNFL pulsation extraction algorithm was proposed to track the displacement of the RNFL which is based on the acquired RNFL B-scan images. Finally, in evaluating the dynamic pulse signal extracted from the pulsating motion of RNFL, an experiment was designed to collect heart rate using an infrared pulse sensor device. The cardiac pulse waveform and the RNFL pulse waveform were compared and analyzed in time and frequency domain. The results show that the extracted RNFL pulse has the same frequency as the cardiac pulse, which validate the feasibility and accuracy of the in vivo extraction scheme used in this paper.