Bandelet Denoising in Image Processing

As digital media and internet use grow, imagery and video are prevalent in many areas of life. Many sensing methods such as Full Motion Video (FMV), Hyperspectral Imagery (HSI), and medical imaging have been developed to accumulate data for diagnostics. Analyzing imagery data to detect and identify specific objects is an essential phase of comprehending visual imagery. Content-based image retrieval (CBIR) is a contemporary development in the field of computer vision. Currently, edge detection filters create undesirable noise for CBIR that leads to difficulties in object detection algorithms. Bandelets have been shown to decrease the noise in signals and images by their use of geometric regularity to compute polynomial approximations in localized regions. In this paper, we use both the bandelet and the discrete wavelet transform to decrease noise within an image. By using Wavelet Exploitation of Bandelet Coefficients (WEBC) to decrease noise we can enhance object detection for CBIR. WEBC raised the peak signal to noise ratio from noised to the denoised images by 19 percent on average, while the structural similarity index measure actually increased by 80 percent on average.