Runtime Performance Enhancement of a Superpixel Based Saliency Detection Model

Reducing computational cost of image processing for various real time computer and robotic vision tasks, e.g. object recognition and tracking, adaptive compression, content aware image resizing, etc. remains a challenge. Saliency detection is often utiliz

  • PDF / 5,352,146 Bytes
  • 11 Pages / 439.37 x 666.14 pts Page_size
  • 88 Downloads / 242 Views

DOWNLOAD

REPORT


)

College of Electrical and Mechanical Engineering, National University of Sciences and Technology (NUST), Islamabad, Pakistan [email protected], [email protected]

Abstract. Reducing computational cost of image processing for various real time computer and robotic vision tasks, e.g. object recognition and tracking, adaptive compression, content aware image resizing, etc. remains a challenge. Saliency detection is often utilized as a pre-processing step for rapid, parallel, bottom-up processing of low level image features to compute saliency map. Subsequent higher level, complex computer vision tasks can then conveniently focus on identified salient locations for further image processing. Thus, saliency detection has successfully mitigated computational complexity of image processing tasks although processing speed enhancement still remains a desired goal. Recent fast and improved superpixel models are furnishing fresh incentive to employ them in saliency detection models to reduce computational complexity and enhance runtime speed. In this paper, we propose use of the superpixel extraction via energy driven sampling (SEEDS) algorithm to achieve processing speed enhancement in an existing saliency detection model. Evaluation results show that our modified model achieves over 60 % processing speed enhancement while maintaining accuracy comparable to the original model. Keywords: Visual saliency · Superpixel · Over-segmentation · SEEDS

1

Introduction

Visual Saliency is a subset of the visual attention mechanism possessed by primates, including human beings, which exercises selective processing of visual stimulus to focus on relatively important and salient parts of the visual scene, leaving the rest unprocessed. It thus enables the brain to cope with the massive influx of sensory visual data (108–109 bits per second), which is otherwise beyond the limited capacity of the brain to handle. Visual attention encompasses both an initial fast pre-attentive stage of bottom-up, low-level, saliency detection based on spontaneous, parallel processing of image data and feature cues (e.g. intensity, color, orientation etc.) and a second attentive stage which is relevance oriented, task dependent, top-down, memory assisted and voluntary selective gaze oriented image processing. The term ‘visual saliency’ is in fact synonymous to bottom-up saliency detection. Over the past two decades much research has been conducted to develop saliency detection models, mostly biologically inspired by the human visual system. Most of such models are of the bottom-up category, starting © Springer International Publishing Switzerland 2016 A. Campilho and F. Karray (Eds.): ICIAR 2016, LNCS 9730, pp. 120–130, 2016. DOI: 10.1007/978-3-319-41501-7_14

Runtime Performance Enhancement

121

with the pioneering work by Itti et al. [1], which is the computational realization of an earlier model based on the renowned feature integration theory (FIT) of attention [2, 3]. The rapid advancement in mobile devices and affiliated computer vision applic