Image Segmentation Using Wavelet Decomposition Method

Image segmentation is the process of partitioning an image into distinct and meaningful regions by organizing pixels with similar characteristics, such as intensity, color, or texture features. This technique is essential in various applications, including medical imaging, autonomous vehicles, and object recognition, as it allows for better analysis and interpretation of visual data. Research on the conventional segmentation methods, such as thresholding or maximum variance, has frequently been faced with noise, fine detail loss, or cross-image compatibility. The K-means clustering technique is employed on the image to reduce the noise. After that, the Morris minimum-bandwidth wavelet transform with second-level decomposition is utilized to enhance image details and extract multi-resolution features into different levels of detail sub-bands. Also, we utilize the Morris minimum-bandwidth wavelet transform with second-level decomposition to decompose the image’s histogram into different levels of detail sub-bands. By focusing on the high-frequency parts of the image, we add them up and identify the densest clusters of data to determine the coarse segmentation areas. Coarse approximations are identified at larger scales, while finer details are preserved at smaller scales to enhance segmentation accuracy. To further refine object boundaries, several edge detection techniques—including Sobel, Canny, Prewitt, Roberts, and zero-crossing—are integrated, followed by the use of an image complement technique to generate the final segmented output. The proposed framework is evaluated on the publicly available BSDS300 dataset, with performance measured using the probabilistic rand index (PRI) against ground-truth references. Experimental results show that the wavelet-based approach achieves competitive segmentation accuracy compared to conventional methods, demonstrating its potential as an effective solution for practical image analysis tasks.

• Supervision