多指标融合的人物图像美感评价算法研究

Research on Image Aesthetic Evaluation Algorithm Based on Multi-Index Fusion

本文针对人物图像,提出一种根据人类视觉及审美习惯来评价其美学价值的可计算方法,从技术特征、感知特征和社会特征3个方面对图像进行特征提取并作为评价指标,融合各指标建立美感评价模型,采用在线调研的方式确定各指标的影响因子,从而实现人物图像美感自动评价.在香港中文大学人物图像美学质量评价数据集(CUHKPQ)上的实验测试结果表明,本文提出的方法准确率可达到98.08%,具有一定的有效性,并且各指标的评价结果清晰直观,可以从各角度对提高图像美学质量起到指导作用.通过将模型集成到手机应用程序中,能够对用户上传的多张人像照片进行美感评价和打分,并依据评分结果选择保留或删除的照片,从而简化用户操作.

This paper proposes a computable method to evaluate the aesthetic value of character images according to human vision and aesthetic habits.Image evaluation indicators are extracted based on technical characteristics,perceptual characteristics and social characteristics.An aesthetic evaluation model is built based on the integration of these indicators.The influence factors of various indicators are determined by online investigation,thus realizing the automatic evaluation of the aesthetic feeling of character images.The experimental results on character image aesthetic quality evaluation data set of CUHK(CUHKPQ)show that the accuracy rate of the proposed method can reach 98.05%,and the evaluation results of each index are clear and intuitive,which can play a guiding role improving the image aesthetic quality from various perspectives.By integrating the model into the mobile phone application program,users can carry out aesthetic evaluation and scoring on a plurality of portrait photos and select the photos retained or deleted the unwanted ones according to the scoring result,thus simplifying the user operation.