| Pollen allergy(hay fever)is a common and frequent disease worldwide,and the prevalence of hay fever has shown a significant upward trend in recent years.Hay fever can have symptoms such as itchy nose,runny nose,sneezing,itchy eyes,and can even cause lower respiratory symptoms such as chest tightness,breath-holding,and asthma.Available survey data show that the prevalence of hay fever in various cities and regions in China ranges from 0.9% to 5%,and even according to the lowest prevalence rate available,the number of hay fever patients in China is large to more than ten million.In Inner Mongolia,there are more plants such as wormwood and dandelion,which make more pollen float during the allergy season,and the trend of hay fever patients is increasing year by year.Understanding pollen shape characteristics through pollen 3D shape reconstruction can help researchers to conduct targeted research on hay fever,and at the same time,it can positively contribute to the popularization of pollen knowledge.To date,researchers have conducted numerous studies on depth image correction,but none of these methods are applicable to pollen image depth map estimation.These methods are targeted at the 3D reproduction of real objects or scenes and are aided by specialized depth information acquisition devices.However,pollen images are generated from the electron reflection signal of pollen and do not possess distance information;Also,the number of pollen samples is small and cannot be estimated by deep learning-based methods.In this paper,we aim at pollen 3D shape reconstruction and study the method of pollen image depth information correction based on image processing.The proposed algorithm for depth information correction of pollen images based on image processing aims to correct the depth information of pollen images on the basis of preserving their original shape information.First,the pollen images are preprocessed and a pollen image longest line extraction method is proposed.To avoid the effect of noise on the resultant image,the image is denoised and then pollen edge information is obtained;Extraction of skeletal edges by refinement of edge images;Finally,the information of the longest line profile of the skeleton edge is calculated.Secondly,a pollen depth image correction method based on tanh function is proposed.The pollen depth images were corrected using the feature that most pollen images are elliptical three-dimensional structures.It is assumed in the study that the grayscale distribution from the centroid to the edge of the pollen image obeys the distribution of the tanh function.The correction principle is that the center of the pollen image is closest to the observation point,so the center of the pollen is corrected by a large amount of gray value.The correction principle is that the center of the pollen image is closest to the observation point,so the grayscale value of the center of the pollen is corrected to a greater extent.The correction principle is that the center of the pollen image is closest to the observation point,so the grayscale value of the pollen center position is corrected to a greater extent.The pollen edge is the farthest from the observation point,so the correction of the grayscale value of the edge is smaller.In this study,the depth information of the background is not considered,so the background of the image is set to black.According to the above principle,the grayscale value of the resulting image decreases from the center of the pollen to the edge,so that the image has depth information visually.Then,two methods of pollen depth map correction based on Gamma function are proposed.In the first method,we modify the distribution obeying the tanh function assumed in the method of pollen depth image correction based on the tanh function to a distribution obeying the Gamma function,and use this as a basis to attach grayscale information obeying the Gamma function to the original image.Make the center point of the corrected image brighter and the edge points remain largely unchanged.However,the pollen edge part of the image corrected by the first method is still brighter,which is not conducive to the 3D shape reconstruction of pollen.Therefore,we further improved the Gamma function-based pollen depth map correction method I and proposed the Gamma function-based pollen depth map correction method II.The graph generated by the Gamma function is the main one when improving.On this basis information about pollen in the original image based on weight values is appended to the graph generated by the Gamma function.The center point has the highest weight and the pollen edge has the lowest weight value when appending grayscale information.Finally,experimental validation of the proposed three methods was carried out.Twodimensional pollen depth images were objectively evaluated using PSNR,SSIM,LOE and other evaluation metrics.The evaluation results showed that both the Gamma functionbased pollen depth map correction method II outperformed the other two methods.Finally,a three-dimensional reconstruction was completed using pollen depth images,and subjective evaluation experiments were conducted.The evaluation method used Thurstone’s pairwise comparison method.The evaluation results further verified that pollen depth map correction method II based on Gamma function was optimal,followed by pollen depth map correction method I based on Gamma function,and the worst result was the pollen depth map correction method based on tanh function. |
