Detection of Placido rings fracture based on ECC image registration

: In this paper, a new detection algorithm for tear film rupture based on Placido ring is proposed based on image registration and region segmentation. The algorithm divides the image into local regions, analyzes the joint entropy of the relevant local image, and judges the correlation between the joint images. The image registration of ECC is a kind of method based on Mutual Information. We judge the correlation between images by analyzing the entropy of relevant images. In this algorithm of ECC, we improve it by dividing the original image into local area regularly. It would improve the contrast of local region. The test results of proposed method can accurately mark the change of Placido rings’ position.


Introduction
Dry eye is a kind of chronic ocular surface diseases in the clinical.Most of people use their eyes for a long time.It becomes the most common reason for the dry eye.At the same time, some diseases of the eyes would also bring the complication of dry eye.Studies have shown that the incidence of dry eyes is increasing.Along with the social aging, the professional of the women and the popularity of cornea contact lens, the diagnosis and treatment of dry eye make everyone pay attention to it [1].
At present, most of the dry eye detection equipment are based on corneal topography.This kind of devices of dry eye is all using the Placido rings for their lighting source.This type of lighting source is extremely special so that it is significant for making up the broken corneal topography.But in this process, we need to fit a series of circular as well as analyze the contour consist of circles.Although the kind of detection for dry eye is already comprehensive, it is too complex and unpractical to the clinical medicine compared with the theoretical study.In clinical practice, we need a more simple and efficient method of detecting dry eye.In this article, we talk about how to use the changes of Placido rings to demarcate the location of broken tear film in direct.

The relationship between the broken placido rings and dry eye
Tears distributed evenly in the conjunctival sac and form a layer of liquid membrane.It is called tear film in medicine name.There are three parts of tear film, including the layer of mucin, water liquid layer and lipid layer.There are some causes for the patients of dry eye that the tear film of dry eye spread unevenly because of the less tears made by dry eye.This kind of tears secretes less than normal eyes, but they evaporate faster.When we light upon the normal eyes using the Placido rings, it would appear a series of whole and perfect rings.On the contrary, we have the same condition to apply on the bad eyes; the projection of Placido rings becomes distorted, eventually broken.In normal conditions, we use the slit lamp to examine dry eye.In this method, we must make the tear film in some specified color.When the tear film changed, we could find the changing out under the slit lamp by watching lost color.In the imaging, the method of Placido rings has the same principle with slit lamp.We merely find out the changing of rings.In addition, the method of using Placido rings would not injure the patient eyes because the tester's eye would not be colored.This new technology is called non-invasive.And non-invasive tear break-up time (NIBUT) has been widely admitted by the medical profession [2].
In ring rupture process, it can easily distinguish the rupture by human eyes.But the tear film appears to cover on the colored cornea, and it appears as the transparent color.Under normal conditions, we cannot see the process of the rupture.After rigorous medical research, it has shown that the rupture of Placido rings has a certain linear relationship with broken tear film.Therefore, we can make the process of rupture of tear film visible by corresponding the changing of Placido rings with the rupture of tear film [3,4].

The theory of image registration based on ECC
ECC is the abbreviation of entropy related coefficient [5].Image registration based on ECC is a member of registration calculated by mutual information.Entropy is an important concept in information theory.Entropy is used to describe the uncertainty measurement system and reflects the amount of information that a system itself can provide.The greater the information entropy of a variable, so does he appeared.The mathematical expression of information entropy is: In the image, we translate the computational expressions of entropy as follows: h i represents the gray image value to 5 of the total number of pixels in image Y.And N represents gray level in image Y.So we can find the probability is where the gray value is i.Gray-level is a shape of measure in representing the gray histogram [6].When the image histogram has one or more of the peak, there is a smaller entropy; Conversely, the greater the entropy value represents that the histogram is flat.
Entropy has special properties [7]: 1.The negative: . The certain events do not carry any information; 3. The symmetry: When all equal to the probability of events, source of entropy is the maximal; 5.The additive properties: The properties fully reflect the knowledge that can be accumulated, on the other hand, they reflect that any complex problem can also be solved step by step.
Joint entropy, reflects the correlation of two random the variables X and Y.If there are two random variables X and Y, X and Y of the joint entropy is expressed as: In this formula, X and Y respectively behave two images.According to the entropy calculation of single image, we can take the advantage of the joint histogram to calculate the joint entropy between them [8].
Define the size of . M, N, respectively, are on behalf of the reference image F and floating image R gray-scale series.
Take reference images and floating image the same coordinates of pixels in the image for a and b.The gray value of a is r and the gray value of b is f.These two gray levels of pixels on forming a new coordinates [r, f].Then calculate the following operations: the two images corresponding to all the pixels.Hist [M, N] matrix, represent two joint gray histogram of the image.Thus, we can calculate the joint probability density of two images [9]: Mutual information (MI), is an important concept in information theory describing the correlation between the two systems [10].MI could also be understood as comprising each case information.Guidelines for the use of mutual information image is to use two images of entropy and joint entropy to reflect the extent of the information contained in each other.For two images X, Y whose mutual information is represented as: As to X as well as Y, the higher the degree of similarity, namely high degree of overlap, the greater its relevance, the smaller the joint entropy and the greater their mutual information MI.
Learn more related content understanding, we shall find the template matching criteria exist FIG mutual information in the original sub-image.
But when the picture is similar in intensity distribution, FIG template may be the same mutual information of a plurality of sub-graph.This makes mutual information error-prone match.Meanwhile, the mutual information of two images is sensitive to the overlap region.If the two images overlapping area is too small, resulting in mutual information is very small, thus reducing our registration accuracy.
To avoid these drawbacks mutual information, mutual information has been improved by Mares:

Based on the local contrast enhancement
Since lighting Placido rings are special, so we need to identify the region as a whole has a certain ring brightness changes.Due to changes in brightness caused by uneven gray scale image, so we set a single gray stretch scope that we lose a lot of gray scale information.In this paper, a regional division of the local gray contrast enhancement methods is raised [11].After matching, we will offset the corresponding pixel gray; the rest will be part of the two images changes.In order to solve the whole gradation changes that are not uniform, we want to focus on the identified region which is divided into a plurality of pixels uniformly small area.In each small area which we were set gray stretch the scope of each region, so that we can change the whole image area more specific manifested enhance local contrast without losing information or change the excessive amplification change information [12].
Figure 1 shows an image at a single threshold range conditions on the global image contrast after stretching.Figure 2 shows the rear area were taken for each individual corresponding threshold range of contrast stretch in all regions revert to the original image composed of images.For comparison, we can clearly find that the single threshold range will lose a lot of gray area of the region changed significantly after contrast stretch process, so that we recognize the annular rupture area has a great influence.After local area contrast stretching, changes in the area of each region can be shown.This improves the accuracy of the detection area of land.

Image registration and identification process
In this paper, testing steps: 1. Image matching based on ECC; 2. Get picture and image matching diagram corresponding changes; 3. We will approximate the curvature of the ring and into the approximate size of the area.The area is divided into 8 equally large rectangle for the convenience of our calculations; 4. Respectively, for each small area contrast stretching, in order to highlight change region; 5. Make binary image, marked change region; 6. Calculate changes in the regional area.Thresholds defined medical screening to calibrate the tear film break area; 7. Mark the screened area on the original.

Experimental verification
In this paper, we make the 400*400 grayscale image which image format is jpg as the original registration and take the 240*240 grayscale image which image format is jpg as a registration template.Ultimately, we cut the registration into two equally large images, as shown on Figure 3 and 4.  Since this image taken in infrared light, we identify the major changes under the black pupil area.So we cut out the main pupil area.We can clearly distinguish the human eye changes in the two images of the ring.We also found that the brightness of the white ring region two images has changed.This is due to the special lighting conditions caused by unavoidable.But the change in position that we need to get there will be more substantial changes in brightness.We will cut a good image corresponding to the pixel positions after registering two subtraction image obtained (shown on Figure 5).In Figure 5 we can see the variation in the amplitude of the gray area by the naked eye.But it is difficult to distinguish changes in specific details.Therefore, by dividing the region, we will need to identify areas were calibrated eight other large rectangular area.We were eight in this area of the rectangle window width and position of the contrast stretch.First, we were to observe the situation of local branches gray area.Figure 6 is a partial rectangular region we selected: The histogram can be seen that the gray area which is mainly concentrated in the range of [0,100].Next, we need to change us into the gray area through a "window frame" box.Here, we choose the box in the grayscale range [37, 42] of the area, and we block the gray level stretch the entire gray scale.We can make changes in the region to identify more clearly:  After a single stretch, we can clearly see the gray-scale changes in the two images.Due to the small changes will not be easy to identify the human eye, at the same time, through the current detecting dry eye study finds more than a certain range when the change in the size of the tear film that is broken to a certain extent before it can be recognized criteria.Therefore, we set a standard threshold value, the removal of an area less than the area threshold value changes.Here we set the threshold value Th 10.By regional standards in line with tear breakup following results:

Conclusion
This paper presents an ECC image registration method based on the detection ring rupture.In the algorithm implementation process, by comparing the local area for analysis and processing, the overall image of the image is divided into several local areas.And to detect changes in the area of the local area, the final summary to the whole global picture of them.This method can avoid the global image intensity distribution is uneven, as well as single threshold range contrast stretch resulting in the loss of an image change information.
, Hist [r, f] represents the number of reference image having a number of gray scale value r and a floating point image with associated gray value of f after traversing Unauthenticated Download Date | 6/22/19 8:33 PM

Figure 1 .
Figure 1.Under a single threshold

Figure 5 .
Figure 5. Difference image after matching

Figure 9 .
Figure 9. Divided rectangular area divided areas, and the contrast stretching results

Figure 10 .
Figure 10.After the removal of less than a threshold area and the nominal position resulting image