Hemorrhage Segmentation and Detection in Retinal Images using Object Detection Techniques and Machine Learning Perspectives

Automated diagnosis of diabetic retinopathy is one of the major goals of hemorrhage detection. Occurrence of hemorrhages in retinal images indicates the early signs of person being suffering from diabetic retinopathy. In this paper, techniques for segmentation and detection of hemorrhages from retinal images are proposed. The objectives of this work primarily focuses of extraction of blood vessel structures and hemorrhages using connected object and Sobel edge detection techniques on green channel extracted from RGB and further classify the local binary pattern features from segmented objects as hemorrhages detected and non-hemorrhage detected images. Pre-processing is performed on retinal images by extraction of green channel images being subject to adaptive histogram equalization, Laplacian filtering followed by morphological bridging operation. Further hemorrhage detection is carried out in both in machine learning and non machine learning approaches. The datasets employed for experimentation is acquired from Indian Diabetic Retinopathy Image Dataset (IDRiD): A Database for Diabetic Retinopathy. A comparative study on the outcomes of both the method is conducted and observed that the results of both methods are in par with each other with promising average accuracy of about 92.31%