AB102.Image blur perception in amblyopia:beyond edges

Background:Understanding the neurophysiological mechanisms of Amblyopia,a neurodevelopmental disorder of the visual cortex,will bring us closer to full recovery.Past findings have been contradictory.Results have shown that despite having severe acuity impairment,amblyopes can nonetheless perceive sharp edges.In this study,we explore the representation of blur through a series of image blur-discrimination and matching tasks,to understand more about the amblyopes’visual system.Methods:Monocular image blur-discrimination thresholds were measured in a spatial two-alternative forced-choice procedure whereby subjects had to decide which image was the blurriest.Subjects also had to interocularly match pictures that were identical to those used for the image blur discrimination task.Ten amblyopes,as well as a group of ten controls were under study.Results:Data on amblyopes and controls will be presented for both experiments.According to previous research that was done on blur-edge discrimination and matching,we predict that subjects’performance will follow a dipper function,that is,all observers will be better at discriminating between both images when a small amount of blur is applied rather than when the image is either sharp or very blurry.We also predict that amblyopes’blur discrimination will be noisier,but that they will paradoxically be able to match the sharpness of the images presented in the matching task.Conclusions:This would confirm our hypothesis about amblyopes’visual system,that they can represent blur levels defined by spatial frequencies that are beyond their resolution limit,and would also raise interesting questions about the visual system in general regarding the different perceptions driven by images versus edges.

Research on single image super-resolution based on very deep super-resolution convolutional neural network

Single image super-resolution(SISR)is a fundamentally challenging problem because a low-resolution(LR)image can correspond to a set of high-resolution(HR)images,while most are not expected.Recently,SISR can be achieved by a deep learning-based method.By constructing a very deep super-resolution convolutional neural network(VDSRCNN),the LR images can be improved to HR images.This study mainly achieves two objectives:image super-resolution(ISR)and deblurring the image from VDSRCNN.Firstly,by analyzing ISR,we modify different training parameters to test the performance of VDSRCNN.Secondly,we add the motion blurred images to the training set to optimize the performance of VDSRCNN.Finally,we use image quality indexes to evaluate the difference between the images from classical methods and VDSRCNN.The results indicate that the VDSRCNN performs better in generating HR images from LR images using the optimized VDSRCNN in a proper method.

Quantification of Blur and Its Applications

Human visual sense has two aspects in our feeling for blurred image, that is, one is the amount of blur depending on object size, the other is the amount of blur independent of the object size. In the former for example, when the image size becomes larger, the author feels smaller amount blur. The quantitative evaluation based on entropy for blurred images is proposed in this paper. The author calls this metric ＂variation entropy＂. This metric has two kinds of aspects that coincide with the human visual sense. The first is the absolute evaluation of blur, and the second is the relative evaluation of blur. The former can be quantified by variation entropy for a unit boundary length （or L-type variation entropy： HL ）, which is dependent on resolution, and the latter can be quantified by variation entropy for a unit area （or A-type variation entropy： H^A ）, which is independent of resolution. These two metrics have complementary properties. At last, two variation entropies are applied to the standard kanji character database, and then the strong relation between variation entropy and accuracy of recognition is discussed. The tendency of writing skills for grades is evaluated by applying the metric to a database collected from school children.

Image blur in high energy proton radiography

Proton radiography has provided a potential development direction for advanced hydrotesting, and its image blur is a crucial point that needs to be deeply studied. In this article, numerical simulation by using the Monte Carlo code Geant4 has been implemented to investigate the entire physics mechanism of high energy proton beam travelling through the object and beamline and arriving at the image plane. This article will mainly discuss the various factors which cause the image blur, including the chromatic aberration of the imaging beamline, the insumcient modulation of an incident particle＇s transverse displacement and angle deviation, the longitudinal length of an object, the influence of containment vessel and otherwise.

Imaging beamline for high energy proton radiography

Proton radiography is a new tool for advanced hydrotesting. This article will discuss the basic concept of proton radiography first, especially the magnetic lens system. Then a scenario of 50 GeV imaging beamline will be described in every particular, including the matching section, Zumbro lens system and imaging system. The simulation result shows that the scenario of imaging beamline performs well, and the influence of secondary particles can be neglected.