Segmenting endoscopic images using adaptive progressive thresholding: a hardware perspective K. Vijayan Asari a , T. Srikanthan b, * a Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA 23529, USA b Centre for High Performance Embedded Systems, School of Computer Engineering, Nanyang Technological University, Nanyang Avenue, Singapore 639798, Singapore Abstract Hardware realization of a novel technique based on adaptive progressive thresholding (APT) for the real-time segmentation of endoscopic images is presented. The APT algorithm is mapped onto a linear array of processing elements with each element of a particular segment communicating with its nearest neighbours. The efficiency and hardware portability of this technique justifies its use in applications that require high performance in real- time. Ó 2002 Elsevier Science B.V. All rights reserved. Keywords: Endoscopic images; Adaptive progressive thresholding; Segmentation 1. Introduction The lumen region and boundary in intestinal images form the preliminary basis of the features usedfornavigationandguidanceinanautomated endoscopy system. The high speed and accurate extractionofthesefeaturesisessentialforreal-time robotic navigation. Since the endoscope uses sev- eral light sources at its tip and the illuminating distancesofthesesourcesarelimited,theintestinal surface lying near the light source will be brighter than the farther ones. Hence, the areas of lowest intensity represent the lumen region in an image. Severalresearchershavepresentedvariousmethods to extract the lumen region from the endoscopic images[1,2].Anadaptiveprogressivethresholding (APT)techniqueforlumenextractionfromagray level endoscopic image is presented in this paper. Due to the excessive processing associated with image processing, many hardware implementation schemesforimageprocessinghavebeenpresented in the literature [3,4]. A pipelined architecture for APTisproposed.ThespecialpurposeVLSIarchi- tectureusesanefficientencodingstrategytoreduce the processing time as well as the communication timeamongthefunctionalmodules. 2. Adaptive progressive thresholding Otsu’s thresholding is based on a discriminant analysis that partitions the image into two classes G 0 and G 1 at gray level ‘t’ such that G 0 ¼f0; 1; Journal of Systems Architecture 47 (2002) 759–761 www.elsevier.com/locate/sysarc * Corresponding author. Tel.: +65-790-6965; fax: +65-792- 0774. E-mail address: astsrikan@ntu.edu.sg (T. Srikanthan). 1383-7621/02/$ - see front matter Ó 2002 Elsevier Science B.V. All rights reserved. PII:S1383-7621(01)00027-3