Image Classification with Adaptive Processing of BSP Image Representation

This paper presents an image classification approach using adaptive processing of Binary Space Partition (BSP) image representation. The method involves collecting Labeled images of buildings, flowers, grasses, and hills, followed by preprocessing steps such as resizing and gray scaling. A BSP Tree is implemented to partition images based on pixel intensity, effectively capturing structural information. Key features, including the number of nodes, leaf nodes, maximum depth, average region size, and intensity statistics, are extracted from the BSP Tree. These features are combined into feature vectors and labeled according to their respective image classes. The dataset is then split into training and testing sets. A Support Vector Machine (SVM) classifier is trained using the feature vectors and labels, and its performance is evaluated using metrics such as accuracy, precision, recall, F1-score, and a confusion matrix. Additionally, visualizations of BSP tree partitions on sample images and the confusion matrix are provided for better interpretability. The proposed method effectively leverages BSP image representation for enhanced feature extraction and classification, demonstrating promising results in categorizing complex image classes.

Index Terms—images Dataset, SVM algorithm, supervised learning, machine learning, image processing.