Microscopic Saliva Image Analysis for Ovulation Detection Using Image Processing and GLCM Features

In women, microscopic saliva imaging plays a crucial role in ovulation phase detection. However, low contrast, uneven illumination and noise in saliva ferning pattern images pose challenges for accurate classification. This paper presents multiple stages of image processing pipeline, including pre-processing, feature enhancement, image reconstruction, segmentation and feature extraction. To ensure uniformity and improve computational efficiency, all images are resized to fixed dimension and then applied Contrast Limited Adaptive Histogram Equalization (CLAHE) for local contrast enhancement. The pre-processed images undergo feature enhancement using Discrete wavelet Transform (DWT) and then Discrete cosine Transform (DCT) in a sequential manner. The enhanced image is then reconstructed using inverse DCT. Unsharp masking is then applied to improve edge clarity. As a next step, Otsu's thresholding and Canny edge detection are applied for segmentation. Furthermore, feature extraction using Gray-Level Co-occurrence Matrix (GLCM) was used to improve texture analysis and produce reliable descriptors for classifying the ovulation and non-ovulation classes. Experimental results demonstrate that, SSIM values which clustered around 0.98, indicating preserved structural integrity. The PSNR values in both cases range from approximately 26 to 46 dB, reflecting high visual quality. The extracted GLCM features helps for the effective classification of images of two classes such that, ovulation images exhibited significantly higher contrast and lower homogeneity compared to non-ovulation images. In addition, lower energy and correlation values in ovulation images suggested reduced uniformity and linear dependency among pixel intensities. The method outperforms traditional enhancement techniques in terms of contrast improvement, feature extraction and segmentation quality. Keywords—Image enhancement, Discrete Cosine Transform, Contrast Limited Adaptive Histogram