The Neighborhood Median Pixel Method (NMPM) classifies land cover by summing per-band scores across 10 input features from Landsat-8 Operational Land Imager (OLI) data – bands 1 through 7 alongside three widely used index images: NDVI, NDWI, and NDBI. These features are typically treated equally within the classification framework, assuming uniform informational value across all bands and indices. However, indices such as NDVI, NDWI, and NDBI are specifically designed to highlight spatial properties of their respective land cover classes—particularly in urban settings—and are therefore expected to carry more relevant information for specific classification tasks. In this study, we experimented with different weighting schemes and found that giving greater emphasis to the indices led to a modest increase in overall classification accuracy, achieving an average overall accuracy of 0.9475 compared to 0.94 of the original implementations for the equal-weight baseline and other weighting strategies with a 9x9 neighborhood size. The results demonstrate how a targeted methodological innovation in image processing—assigning greater weight to highly relevant features—can enhance the reliable and efficient classification performance of the NMPM. This contributes to more accurate land cover mapping, particularly in complex urban environments, and supports data-driven development planning and resource management.

The Neighborhood Median Pixel Method has previously been introduced as an image processing technique in remote sensing, developed to classify Landsat-8 OLI satellite image pixels into categories of vegetation, water, and built-up areas. This method relies on a lookup table based on the median pixel values within a pixel’s neighborhood and a scoring system that assigns point values for classification. While a 9x9 neighborhood size was originally proposed, a succeeding study suggested a 13x13 neighborhood for better classification accuracy. This study focuses on refining the scoring system used in the Neighborhood Median Pixel Method, particularly the original set of arbitrary point values 14, 4, and 1. Particle Swarm Optimization was employed to systematically explore and optimize these point values, iteratively seeking an ideal configuration for the scoring system. After optimization, the Neighborhood Median Pixel Method exhibited a slight increase in overall accuracy. Using the 9x9 neighborhood size, the accuracy rose from 94% to 94.75%, while with the 13x13 neighborhood size, the accuracy improved from 95.75% to 96%. Furthermore, results indicate that varying point value configurations yield similar classification outcomes, suggesting that the method’s scoring system is robust across multiple configurations and that the originally proposed point value set remains adequate for effective classification.

Dental shade matching for composite restoration to natural teeth color is a crucial aspect of dental treatment as it can significantly impact patient satisfaction and treatment outcomes. However, the subjective nature of manual shade selection often leads to shade mismatch, which leads to failure on the first visit. In addition, intraoral scanners are inaccessible to small enterprises dental clinic in the Philippines due to its unaffordable pricing. To address this problem, this study proposed a mobile application that utilizes image processing and deep learning techniques for objective and consistent dental shade matching. Exploring Convolutional Neural Network (CNN)-based MediaPipe for Facial Landmark Detection and Support Vector Machines (SVMs) to classify dental shades. The SVM model attained an overall accuracy of 68.5% during the experimental results while the implementation using the mobile application obtained an estimate of 90% during the user testing for A1 to A4 color shade. The findings have significant implications for clinical practice, empowering dental professionals with a reliable tool to improve patient care and satisfaction. This study emphasizes the importance of incorporating advanced technology into clinical practice, ultimately improving patient outcomes.

Optical character recognition (OCR) is the process of extracting the characters from a digital image. The concept behind OCR is to acquire a text in a video or image formats and extract the characters from that image and present it to the user in an editable format. In this study, a convolutional neural network (CNN) is applied, which is a mathematical representation of the functionality of the human brain, using back-propagation algorithm with test case files of English alphabets and numbers. The purpose of this study is to test systems capable of recognizing vehicle plate number English alphabets and numbers with different fonts, and to be familiar with CNN and digital image processing applied for character recognition. Scientific journals and reports were used to research the relevant information required for the thesis project. The chosen software was then trained and tested with both computer and video output files. The tests revealed that the OCR software can recognize both vehicular plate and computer alphabets and learns to do it better with each iteration. The study shows that although the system needs more training for vehicular plate characters than computerized fonts, and the use of CNN in OCR is of great benefit and allows for quicker and better character recognition.

Complete Blood Count is one of the most commonly performed medical laboratory procedure today. It is required to detect various types of diseases. Presently, some small-scale clinics in the country still does the tedious, manual method of counting the blood cell. With Complete Blood Count Analysis System through Image Processing, automated CBC can be performed by mounting the smart phone camera on the viewer of the microscope. The input image will go through several image processing algorithms such as: Binary Thresholding, Clustering, and Hough Circle Technique. The result will be computed through the formulas used in the manual method of the CBC process. Experimental results show the developed system gains 94% of accuracy for counting the Hematocrit, Hemoglobin, Red Blood Cell, and White Blood Cell values.