The proposed methodology of this paper is designed based on the dual strategic Machine Learning principles such as Depth Optimized Machine Learning Strategy (DOMLS) and the Modified K-Means Optimization Logic (MkMOL). This paper introduced a new algorithm called DOMLS, in which it integrates all the digital image processing constraints together to provide an efficient solution to identify the Glaucoma disease on retinal area as well as reduce the time complexity ratio in good manner. The proposed logic is useful to identify the glaucoma disease on earlier stages and provide proper predictions accordingly with respect to the machine learning strategies. This paper associates many techniques together to produce an efficient machine learning and optimization strategy called DOMLS and MkMOL. Based on the following digital image processing associations to identify the glaucoma disease are as follows: Image Acquisition and ROI Selection, Optical-Disc Segmentation Process, Hue, Saturation, Lightness (HSV) Plane Enhancement, Optimization Logic Association and Classification. All these details are illustrated in detail as below.
A. Image Acquisition and ROI Selection
The input image acquisition process begins with the selection of OCT images from the image repository and the process of Region-of-Interest (ROI) selection depends on the accuracy factor. In which the ROI selection process allows the user to select the specific region of input to process further accordingly. The general formation of ROI selection is in rectangle format, in which it is easy to acquire only the selected part and apply the feature processing logic only to that extracted part for improving the accuracy levels in outcome. The following algorithm, Algorithm-1 illustrates the logic of image acquisition and ROI selection in clear manner with proper Pseudocode specification.
Algorithm-1: Image Acquisition and ROI Selection
Input: OCT Image for Processing.
Output: Return the ROI Segmented Portion.
Step-1: Flush out all the cache images and accumulate the input image from the respective system..
Step-2: Define the variable to acquire the image with specific format inclusions.
Step-3: Use the function for reading the image called "imread" to obtain the input image into the created object on Step-2.
Step-4: Crop the selected OCT image by using image cropping function called "imcrop".
Step-5: Show the Cropped portion of the image to the user to select the respective ROI area from that noise free image.
Pseudocode:
clear all_cache;
[F_name, p_name]=UI_get_file[{'*.jpg;*.png;*.tif'},'Image_File_Formats'];
Def I=imread(p_name F_name);
I=imresize(I,[256 256]);
out=imcrop(I); img_show(out);
Step-6: Analyze the Red, Green and Blue (RGB) color portions of the ROI selected and processed image.
Pseudocode:
Red_plane=out[:,:,1]; Green_plane=out[:,:,2]; Blue_plane=out[:,:,3];
Step-7: Plot the selected portions to the user view for identification and validations.
Pseudocode:
Subplot(1);image_show[out];image_head['ROI'];
Subplot(2);image_show[Red_plane];image_head['Red_Palne'];
Subplot(3);image_show[Green_plane];image_head['Green_Plane'];
Subplot(4);image_show[Blue_plane];image_head['Blue_Plane'];
ROI_img=image[Red_plane, Green_plane, Blue_plane];
Step-8: Return the ROI Selected image for further processing.
Pseudocode:
return ROI_img;
B. Optical-Disc Segmentation Process
Segmenting the optical disc is a significant and basic advance in making a casing of reference for diagnosing optical nervous system associations, for example, Glaucoma. In this manner, a solid optical disc segmentation procedure is vital for programmed screening of optical nervous system irregularities. The principle commitment of this paper is in introducing a novel optical disc division procedure dependent on integrating a sequential process on a restricted optical disc image. To keep the veins from meddling with the sequential cycle, an in painting strategy is utilized. Too a significant commitment is to include the varieties in decisions among the Ophthalmologists in distinguishing the optical disc limits and diagnosing the glaucoma. The vast majority of the past summaries and investigations were prepared and tried dependent on just a single assessment, which can be thought to be one-sided for the Ophthalmologist. Along with this the proper disc segmentation leads to higher accuracy rates in result and provides the low iteration level in outcome. The following algorithm, Algorithm-2 illustrates the logic of Optical-Disc Segmentation Process in clear manner with proper Pseudocode specification.
Algorithm-2: Optical-Disc Segmentation Process
Input: ROI extracted image from Algorithm-1.
Output: Return the segmented image with associated features.
Step-1: Create an object named 'im" and Read the ROI extracted image.
Pseudocode:
im =Red_plane;
Step-2: Create 2 unique objects named 'cl2 and cl3" and to accumulate the Green and Blue plane values.
Pseudocode:
cl = imclose(im, ones(9));
cl2 = imclose(Green_plane, ones(9));
cl3 = imclose(Blue_plane, ones(9));
Step-3: Accumulate the coefficients of three planes into the respective array variables.
Pseudocode:
co(:,:,1)=cl; co(:,:,2)=cl2; co(:,:,3)=cl3;
Step-4: Sharpen the estimated image with respect to created coefficients.
Pseudocode:
C2shap=co; C2shap(:,:,2)=200;
ImShar=imsharpen(C2shap);
Step-5: Segmenting the sharpen images based on Step-4.
Pseudocode:
Define Img_Segment;
Img_Segment=Segment(co{1,1,1})[ImShar];
Step-6: Return the Optical Disc Segmented image to the HSV plane enhancement stage.
Pseudocode:
return Img_Segment;
C. Hue, Saturation, Lightness (HSV) Plane Enhancement
During the Red, Green and Blue shading model alludes to the handling of color-tones in the manual visual framework, the HSV color model compares to the human impression of color likeness. In this paper, a projection of RGB vectors inside the RGB color space is figured out, in which it isolates non-chromatic from chromatic data. The projection is the manipulation which is equal to Hue and Saturation of the HSV shading space in the RGB color portion. It incorporates the psycho visual idea of human separation between colors of the HSV space into the physiological visual based idea of the RGB space. With the perception of it is, as opposed to the overarching assessment, conceivable to separate between colors dependent on human discernment in the straight math of the RGB color space. This opens additional opportunities in numerous fields of color image handling, particularly in the area of Glaucoma based image processing schemes. The following algorithm, Algorithm-3 illustrates the model of HSV image plane enhancement feature construction process in detail with proper Pseudocode specifications.
Algorithm-3: HSV Plane Enhancement
Input: Extract the image features from Algorithm-3.
Output: Enhanced Image with HSV features.
Step-1: Create four individual objects called hsv, h, s and v, in which it obtains the HSV values that is converted from RGB image features.
Pseudocode:
hsv=rgb2hsv(ImShar);
h=hsv(:,:,1);s=hsv(:,:,2);v=hsv(:,:,3);
Step-2: Display all the channel values to the user perspective.
Pseudocode:
figure();subplot(1);img_show(hsv);
subplot(1);
img_show(v);
Step-3: Convolute the image features with respect to 'adapthisteq()' equalizer.
Pseudocode:
v=adapthisteq(v);
Step-4: Create an object called 'Disc_Image' to store the concatenation values of the HSV proportions.
Step-5: Enhancing the proportions of image and display it to the user end.
Pseudocode:
cEnhance=zeros[size(h,1),size(h,2),3];
cEnhance(:,:,1)=h; cEnhance(:,:,2)=s; cEnhance(:,:,3)=v;
img_show[{hsv Disc_Image cEnhance}];
Step-6: Return the enhanced plane image for optimization.
Pseudocode:
return Disc_Image;
D. Optimization Logic Association and Classification
The proposed methodology of DOMLS adapts the new methodology of optimizer called Modified K-Means Optimization Logic (MkMOL), in which the optimizer performs the cluster based glaucoma image optimization. The necessity of cluster based approach is utilized over this paper because of attaining the higher accuracy ranges with lesser error ratio. The best optimum results will be gathered on each level of cluster considerations and the vector points of each set glaucoma images are considered as a global mean and process accordingly. The general k-means cluster logic is enhanced over this machine learning process and improves the accuracy by means of adding some new convolutions and shaping natures. The following algorithm, Algorithm-4 illustrates the logical flow of proposed optimization process as well as the scenario indicates the dual optimization flow with respect to HSV constructed images, so that the enhanced form of k-means logic is manipulated over this approach. The structured form of HSV images are reconstructed based on best possible cluster region outcome and the local optimization features of such process will improve the quality of prediction in results. These regions will be segmented by using segmentation principles. The classification model identifies those segmented glaucoma image features and process the matching optimized regions based on the training set and produces the exact classification summary with respect to proposed approach of DOMLS.
Algorithm-4: Optimization and Classification
Input: Obtain the HSV Plane Enhanced Image features from Algorithm-3.
Output: Classified Results with Accuracy ratio.
Step-1: Create an object called cform to store the RGB color lab library functions.
Step-2: Apply the image enhancement ratio to the created object cform on Step-1 and stored that into the new object named lab_hsv.
Step-3: Initiate the dual convolution process to optimize the layers of the lab_hsv.
Step-4: Identify the number of rows and columns presented into the dual convoluted cform object with respect to lab_hsv.
Pseudocode:
cform=make_c_form('srgb2lab');
lab_hsv=apply_cform(c_Enhance,cform);
ab=dual(lab_hsv(:,:,2:3));
n_rows = size(ab,1);
n_cols = size(ab,2);
Step-5: Reshape the generated row and column matrix and store it into the object called ab.
Pseudocode:
ab=reshape[ab,n_rows*n_cols,2];
Step-6: Define a variable called ncolors and assign the constant value to it.
Pseudocode:
ncolors=2;
Step-7: Create a cluster index with respect to the color ranges, row and column index ratio, euclidean specifications and the associated replications.
Pseudocode:
[cluster_index, cluster_size] = enhancekm(kmeans[ab,nColors,'distance','sqEuclidean', 'Replicates',3]};
Step-8: Reshape the image indexing based on the created cluster index values (Step-7) based on the defined matrices.
Step-9: Assign the respective labels to the generated pixels with respect to RGB color index scheme.
Step-10: Display the resulting images to the user end.
Pseudocode:
pixel_labels=reshape[cluster_index,n_rows,n_cols];
Pixel_labels=~[pixel_labels>1];
img_show[Pixel_labels];
Step-11: Rasie the classification principle to cross check the finalized image into the dataset features to check the given image is considered to be Glaucoma or not as well as the accuracy enumeration is handled through the generated classification law.
Pseudocode:
Check if input(pixel_labels==dataset(k));
segmented_images{k}=color;
Acc_enum=str2num(Dec{segmented_images{k}});
Step-12: Display the classified image category and raise the alert to indicate the testing image is Glaucoma or not with proper accuracy levels.
Pseudocode:
img_show(segmented_images{k});
return Acc_enum;