论文部分内容阅读
1 Introduction
Imaging fiber bundles have been attracting great interest for their effectiveness as imaging probes for high-resolution optical imaging.But the resulting lateral image is not a continuous image but contains a pixelated artifact image is not a continuous image but contains a pixelated artifact resembling a honeycomb.This study focus on eliminate the honey comb structures.
2 Proposed methods and results
The main idea of this study is that similar pixelated images are response to similar depixelized image.An external pixelated image dictionary is used to reconstruct any other input pixelated image，and at the same time a depixeled image is construct with the matched elements in the dictionary pairs.
2.1 Obtain the external dictionary
2.1.1 Simulate the pixelized image
We Generated the synthetic pixelated image.A mirror M image，which is obtained in fiber optic system without any object is used for generating the pixelated image.The simulated pixelated image is calculated with over-lapping the mirror image to the depixelized image H that ： F= H.*M，where.* is the multiply operation for each pixel values in the two image at matching location.
2.1.2 Obtain the dictionary
Image patches are extracted from the image pairs with patch size ps over lapping size pov，with the direction top to down，left to right.
The patch size are decided depend on the size of the fiber cores，the over lapping size is the larger the better.
This process is carried twice，once for pixelated image，once for depixelated image so we could obtain a dictionary with one to one mapping patches.
2.2 Extract the patches
Do this process the same way with obtaining the dictionary.Pli is the extracted patches from image.
Represent the patches of the under depixelized image with a most similar element in the dictionary：find dln∈Dl，minP.
2.3 Reconstruct the under depixelized image
Reconstruct the under depixelized image with matched combination of dictionary patches.The proposed depixelized image is also reconstructed at the same time：
Phi≈dhn，dhn is the paired patch of dln in the dictionary Dh.
2.4 Experiment results
The fiber bundle diameter is 6 pix in the image;fibers gap is 3 pixel/direction，links two centroid of fiber bundle cores;patch size is 7*7pixel， which means square patches are extracted，overlapping size are 5*5 pixel.
3 Conclusion
In summary，we developed an exampler-based depixelized method with external dictionary for eliminating the pixelation effect for the structural artifact from endoscopic fiber bundle images.we successfully removed the artifacts and obtained a depixelized image.
Reference
[1]Depixelation of coherent fiber bundle endoscopy based on learning patterns of image prior.
[2]W.T.Freeman Exampler -based super-resolution.
收稿日期：2018-4-14
Imaging fiber bundles have been attracting great interest for their effectiveness as imaging probes for high-resolution optical imaging.But the resulting lateral image is not a continuous image but contains a pixelated artifact image is not a continuous image but contains a pixelated artifact resembling a honeycomb.This study focus on eliminate the honey comb structures.
2 Proposed methods and results
The main idea of this study is that similar pixelated images are response to similar depixelized image.An external pixelated image dictionary is used to reconstruct any other input pixelated image，and at the same time a depixeled image is construct with the matched elements in the dictionary pairs.
2.1 Obtain the external dictionary
2.1.1 Simulate the pixelized image
We Generated the synthetic pixelated image.A mirror M image，which is obtained in fiber optic system without any object is used for generating the pixelated image.The simulated pixelated image is calculated with over-lapping the mirror image to the depixelized image H that ： F= H.*M，where.* is the multiply operation for each pixel values in the two image at matching location.
2.1.2 Obtain the dictionary
Image patches are extracted from the image pairs with patch size ps over lapping size pov，with the direction top to down，left to right.
The patch size are decided depend on the size of the fiber cores，the over lapping size is the larger the better.
This process is carried twice，once for pixelated image，once for depixelated image so we could obtain a dictionary with one to one mapping patches.
2.2 Extract the patches
Do this process the same way with obtaining the dictionary.Pli is the extracted patches from image.
Represent the patches of the under depixelized image with a most similar element in the dictionary：find dln∈Dl，minP.
2.3 Reconstruct the under depixelized image
Reconstruct the under depixelized image with matched combination of dictionary patches.The proposed depixelized image is also reconstructed at the same time：
Phi≈dhn，dhn is the paired patch of dln in the dictionary Dh.
2.4 Experiment results
The fiber bundle diameter is 6 pix in the image;fibers gap is 3 pixel/direction，links two centroid of fiber bundle cores;patch size is 7*7pixel， which means square patches are extracted，overlapping size are 5*5 pixel.
3 Conclusion
In summary，we developed an exampler-based depixelized method with external dictionary for eliminating the pixelation effect for the structural artifact from endoscopic fiber bundle images.we successfully removed the artifacts and obtained a depixelized image.
Reference
[1]Depixelation of coherent fiber bundle endoscopy based on learning patterns of image prior.
[2]W.T.Freeman Exampler -based super-resolution.
收稿日期：2018-4-14