What Should We Learn from... Bio-Inspired Systems (I)
Bio-inspired systems can be traced back to the invention of radar systems. They are getting even more attention in recent years with topics such as the internet-of-things, intelligence and cognition becoming more and more relevant. Actually, people never stop the pave of learning from nature, animals and human beings.
Acoustic signals used by dolphins are studied to develop sonar systems in marine applications[1]. Bats’ multi-frequency transmission systems afford us lessons to optimize the cognitive radar systems for echolocation, tracking and detection targets[2]. A tree-inspired patch antenna array is proposed to improved the antenna’s gain by selecting the phase of the source feeding each element[3]. On the other hand, visualization plays more and more important roles in signal processing systems especially at the stage of decision. As the image or even video can be regarded as the detection plane, vision system inspired techniques are growing rapidly. The advanced algorithm of image or video processing with intelligent characteristics generally oriented from mammals or human being’s vision systems. The question is What are we looking for when viewing a visual scene?
Recent studies suggest that the answer to this question could be revealed via statistical analysis of human eye fixations. One possible answer named super Gaussian component was investigated in recent published paper “Towards Statistical Modeling of Saccadic Eye-movement and Visual Saliency”[4].
In this paper, the authors present a unified statistical framework for modeling both saccadic eye movements and visual saliency. By analyzing the statistical properties of human eye fixations on natural images, they found that human attention is sparsely distributed and usually deployed to locations with abundant structural information. Based on such observations, the authors propose a novel strategy to model saccadic behavior and visual saliency jointly based on Super Gaussian Component (SGC) analysis. The proposed model sequentially obtains super Gaussian components using iterative projection pursuit, and generates eye-movements by selecting the location with maximum SGC response. Besides human saccadic behavior simulation, the paper also demonstrated the superior effectiveness and robustness of the SGC model over state-of-the-arts by carrying out dense experiments on synthetic patterns and human eye fixation benchmarks.
Multiple key issues in saliency modeling research, such as individual differences, the effects of scale and blur, are also explored in this work. In model adaptively experiment, similar image-level performances among the tested models have been observed, and the degeneration caused by inter-viewer inconsistency was also discovered. The experiments on the effects of scale show that the multi-scale SGP scheme outputs overall better results, yet it still cannot beat the best single scale scheme in 65% of the cases, because the salient patterns in these cases are mostly gathered in a certain scale. Experiments on the effects of feature dimensions show that increasing the number of features might not always improve the model's overall performance and using a complete feature representation is usually not the optimal solution.
Different from previous works that mostly aim to reproduce the exact mechanisms of visual perception, inspirations are drawn from the statistical characteristics of real-world human behavior. Based on extensive qualitative and quantitative experimental results, the paper shows promising potentials of statistical approaches for both human behavior research and signal processing applications.
In this paper, the authors present a unified statistical framework for modeling both saccadic eye movements and visual saliency. By analyzing the statistical properties of human eye fixations on natural images, they found that human attention is sparsely distributed and usually deployed to locations with abundant structural information. Based on such observations, the authors propose a novel strategy to model saccadic behavior and visual saliency jointly based on Super Gaussian Component (SGC) analysis. The proposed model sequentially obtains super Gaussian components using iterative projection pursuit, and generates eye-movements by selecting the location with maximum SGC response. Besides human saccadic behavior simulation, the paper also demonstrated the superior effectiveness and robustness of the SGC model over state-of-the-arts by carrying out dense experiments on synthetic patterns and human eye fixation benchmarks.
Multiple key issues in saliency modeling research, such as individual differences, the effects of scale and blur, are also explored in this work. In model adaptively experiment, similar image-level performances among the tested models have been observed, and the degeneration caused by inter-viewer inconsistency was also discovered. The experiments on the effects of scale show that the multi-scale SGP scheme outputs overall better results, yet it still cannot beat the best single scale scheme in 65% of the cases, because the salient patterns in these cases are mostly gathered in a certain scale. Experiments on the effects of feature dimensions show that increasing the number of features might not always improve the model's overall performance and using a complete feature representation is usually not the optimal solution.
Different from previous works that mostly aim to reproduce the exact mechanisms of visual perception, inspirations are drawn from the statistical characteristics of real-world human behavior. Based on extensive qualitative and quantitative experimental results, the paper shows promising potentials of statistical approaches for both human behavior research and signal processing applications.
References:
[1] Y. Pailhas C. Capus K. Brown. Y. Pailhas C. Capus K. Brown. Dolphin-inspired sonar system and its performance. IET Radar Sonar Navig., 2012, 6(8), pp. 753–763 753
[2] Simon Haykin. Cognitive Radar: A way of the future. IEEE Signal Processing Magazine. 2006, 23(1) , pp. 30-40[3]Karl F. Warnick. A Bio-Inspired Patch Antenna Array Using Fibonacci Sequences in Trees. IEEE Antennas and Propagation Magazine, 2013, 55(5), pp. 192-201
[4] Xiaoshuai Sun, Hongxun Yao, Rongrong Ji, Xian-Ming Liu. Towards Statistical Modeling of Saccadic Eye-movement and Visual Saliency. IEEE Transactions on Image Processing. 2014, 23(1), pp. 4649-4662.
