A reconfigurable architecture for object detection using adaptive threshold

Abstract

The detection of objects is important in many computer vision applications. This paper proposes a reconfigurable architecture for object detection using adaptive threshold with an efficient algorithm for removal of salt and pepper noise from the colour and grayscale images. The main objective of this paper is to design an alternate architecture of object detection using adaptive threshold. In this paper, a type median filter is used to preserve the edges and to reduce the salt and pepper noise easily of the input and reference image is discussed. The pre-processed images are applied to 2D-discrete wavelet transform (2D-DWT) to remove variable illumination and to select appropriate sub-band, i.e., low-low (LL) band which contains maximum information of the original image. The modified background subtraction is used to remove the background from LL band of input and reference images to obtain a foreground image. The detected object is fed to median filter to remove any small amounts of noise which is still present in the image. The modified decision based partially trimmed global median (MDBPTGM) filter was used to give better results in terms of mean square error (MSE), peak signal to noise ratio (PSNR) and image enhancement factor (IEF). Hardware parameters such as slice registers and flip-flop pairs, latches, lookup table (LUT), shift registers and memory usage were compared with the existing techniques. Propose architecture used less number of hardware parameters. It means the proposed design reduces power and the area usage in comparison to the other techniques.