Image Complexity Guided Network Compression for Biomedical Image Segmentation

Compression is a standard procedure for making convolutional neural networks (CNNs) adhere to some specific computing resource constraints. However, searching for a compressed architecture typically involves a series of time-consuming training/validation experiments to determine a good compromise between network size and performance accuracy. To address this, we propose an image complexity-guided network compression technique for biomedical image segmentation. Given any resource constraints, our framework utilizes data complexity and network architecture to quickly estimate a compressed model which does not require network training. Specifically, we map the dataset complexity to the target network accuracy degradation caused by compression. Such mapping enables us to predict the final accuracy for different network sizes, based on the computed dataset complexity. Thus, one may choose a solution that meets both the network size and segmentation accuracy requirements. Finally, the mapping is used to determine the convolutional layer-wise multiplicative factor for generating a compressed network. We conduct experiments using 5 datasets, employing 3 commonly-used CNN architectures for biomedical image segmentation as representative networks. Our proposed framework is shown to be effective for generating compressed segmentation networks, retaining up to ≈95% of the full-sized network segmentation accuracy, and at the same time, utilizing ≈32x fewer network trainable weights (average reduction) of the full-sized networks.

What's in a Photograph? The Perspectives of Composition Experts on Factors Impacting Visual Scene Display Complexity for Augmentative and Alternative Communication and Strategies for Improving Visual Communication

Purpose Visual scene displays (VSDs) can support augmentative and alternative communication (AAC) success for children and adults with complex communication needs. Static VSDs incorporate contextual photographs that include meaningful events, places, and people. Although the processing of VSDs has been studied, their power as a medium to effectively convey meaning may benefit from the perspective of individuals who regularly engage in visual storytelling. The aim of this study was to evaluate the perspectives of individuals with expertise in photographic and/or artistic composition regarding factors contributing to VSD complexity and how to limit the time and effort required to apply principles of photographic composition. Method Semistructured interviews were completed with 13 participants with expertise in photographic and/or artistic composition. Results Four main themes were noted, including (a) factors increasing photographic image complexity and decreasing cohesion, (b) how complexity impacts the viewer, (c) composition strategies to decrease photographic image complexity and increase cohesion, and (d) strategies to support the quick application of composition strategies in a just-in-time setting. Findings both support and extend existing research regarding best practice for VSD design. Conclusions Findings provide an initial framework for understanding photographic image complexity and how it differs from drawn AAC symbols. Furthermore, findings outline a toolbox of composition principles that may help limit VSD complexity, along with providing recommendations for AAC development to support the quick application of compositional principles to limit burdens associated with capturing photographic images. Supplemental Material https://doi.org/10.23641/asha.15032700

Color Texture Image Complexity—EEG-Sensed Human Brain Perception vs. Computed Measures

In practical applications, such as patient brain signals monitoring, a non-invasive recording system with fewer channels for an easy setup and a wireless connection for remotely monitor physiological signals will be beneficial. In this paper, we investigate the feasibility of using such a system in a visual perception scenario. We investigate the complexity perception of color natural and synthetic fractal texture images, by studying the correlations between four types of data: image complexity that is expressed by computed color entropy and color fractal dimension, human subjective evaluation by scoring, and the measured brain EEG responses via Event-Related Potentials. We report on the considerable correlation experimentally observed between the recorded EEG signals and image complexity while considering three complexity levels, as well on the use of an EEG wireless system with few channels for practical applications, with the corresponding electrodes placement in accordance with the type of neural activity recorded.

Preparatory Experiments Regarding Human Brain Perception and Reasoning of Image Complexity for Synthetic Color Fractal and Natural Texture Images via EEG

Texture plays an important role in computer vision in expressing the characteristics of a surface. Texture complexity evaluation is important for relying not only on the mathematical properties of the digital image, but also on human perception. Human subjective perception verbally expressed is relative in time, since it can be influenced by a variety of internal or external factors, such as: Mood, tiredness, stress, noise surroundings, and so on, while closely capturing the thought processes would be more straightforward to human reasoning and perception. With the long-term goal of designing more reliable measures of perception which relate to the internal human neural processes taking place when an image is perceived, we firstly performed an electroencephalography experiment with eight healthy participants during color textural perception of natural and fractal images followed by reasoning on their complexity degree, against single color reference images. Aiming at more practical applications for easy use, we tested this entire setting with a WiFi 6 channels electroencephalography (EEG) system. The EEG responses are investigated in the temporal, spectral and spatial domains in order to assess human texture complexity perception, in comparison with both textural types. As an objective reference, the properties of the color textural images are expressed by two common image complexity metrics: Color entropy and color fractal dimension. We observed in the temporal domain, higher Event Related Potentials (ERPs) for fractal image perception, followed by the natural and one color images perception. We report good discriminations between perceptions in the parietal area over time and differences in the temporal area regarding the frequency domain, having good classification performance.