基于视觉神经元ON-OFF模型的图象增强

Image Enhancement Based on the ON-OFF Model of Visual Neurons

探讨基于视觉神经网络 ON- OFF模型的图象增强 .通过计算仿真 ,找到了关于这种视觉模型应用于图象处理合适的实现形式 ;探讨了衰减常数和颜色恒定性的关系 ;以及空间常数变化和动态范围压缩与细节增强能力的联系 .使用合适的衰减常数 ,可以使该神经元模型在颜色恒定性计算上性能优良 ,同时 ,适当大小的空间常数 ,能够在颜色保真度和图像增强性能之间取得合适的平衡 .神经网络对复杂背景图象的增强效果良好 ,但是由于同样基于“灰度世界”假设 ,因此在处理违反这一假设的特殊图象时 ,此模型在颜色表征上仍然具有缺陷 ,这就部分限制了模型的应用 .最后探讨了可能的改进方向 .

This paper discusses an algorithm of color image enhancement that is based on a neuro-dynamical model of the ON-OFF neurons in the human visual system. An appropriate form of this model for color image processing is found and the advantages and disadvantages of this model are also analyzed through the computational simulations. Extensive computations demonstrate that this model can achieve a very good degree of color constancy with the selection of a proper passive decay rate constant. At the same time, the trade-off between image enhancement and the fidelity of chromatic rendition is determined by the space surround constant. This neural system performs well on the enhancement for the natural scenes with complex contexts. However, because this neural system is a model of receptive fields of the ganglion cells in the human retina and still based on “gray-world” assumption, and it fails to handle the violations of the gray-world assumption. It shows that this model is still not comprehensive enough to describe the complex visual system and have some restrictions in practical application. Finally, we discussed the possible future improvement of this model.