Scholarlink

<p>Image aesthetic assessment (IAA) is a pivotal task in computer vision, with broad applications in intelligent image processing and recommendation systems. However, most existing IAA methods are based on artificial neural networks (ANNs), which suffer from high computational complexity and energy consumption, limiting their deployment in real-time scenarios. To address these challenges, this paper proposes a spiking neural network (SNN)-based approach for image aesthetic assessment that integrates color perception with spatiotemporal deep features. Specifically, we design a learnable integer leaky integrate-and-fire (LI-LIF) neuron model to alleviate quantization errors and introduce spatiotemporal convolution to enable effective information interaction across different time steps. In addition, a spiking-based color distribution attention network is proposed to enhance color-aware aesthetic perception. Extensive experiments conducted on the AADB, PARA, and KonIQ-10K datasets demonstrate the effectiveness of the proposed method. In particular, the proposed model achieves an energy efficiency that is 2.495 times higher than that of equivalent ANN-based architectures on the AADB dataset, highlighting its potential for real-time and energy-constrained applications. The code will be released at <a href="https://github.com/liuxida/SICF-Net">https://github.com/liuxida/SICF-Net</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

SNN-driven aesthetic assessment: integrating color perception with spatiotemporal deep features

  • 期刊论文

  • 发表日期：

    2026年5月17日

查看全文
  • Yang Liu,
  • Jingwen Liu,
  • Huaxu He,
  • Kun Cai

摘要

Image aesthetic assessment (IAA) is a pivotal task in computer vision, with broad applications in intelligent image processing and recommendation systems. However, most existing IAA methods are based on artificial neural networks (ANNs), which suffer from high computational complexity and energy consumption, limiting their deployment in real-time scenarios. To address these challenges, this paper proposes a spiking neural network (SNN)-based approach for image aesthetic assessment that integrates color perception with spatiotemporal deep features. Specifically, we design a learnable integer leaky integrate-and-fire (LI-LIF) neuron model to alleviate quantization errors and introduce spatiotemporal convolution to enable effective information interaction across different time steps. In addition, a spiking-based color distribution attention network is proposed to enhance color-aware aesthetic perception. Extensive experiments conducted on the AADB, PARA, and KonIQ-10K datasets demonstrate the effectiveness of the proposed method. In particular, the proposed model achieves an energy efficiency that is 2.495 times higher than that of equivalent ANN-based architectures on the AADB dataset, highlighting its potential for real-time and energy-constrained applications. The code will be released at https://github.com/liuxida/SICF-Net.