CAJ | 학술논문

随着计算机视觉研究的不断进步和图像超分辨率技术的不断完善，人脸检测图像的分辨率增强技术已然成为人们关注的热点。本文基于人脸检测图像分辨率增强的研究现状和传统Retinex算法的缺陷问题，引入几何包络原理和多分辨率算法，提出了针对人脸检测图像分辨率增强的改进Retinex算法。该算法在输入图像上直接求得一层几何包络作为照度图像的估计，继而求得景物反射系数，并且采用多分辨率算法中的图像金字塔原理进行求解全局性的优化。实验结果表明，本文提出的改进算法较传统Retinex算法，在人脸检测图像的分辨率增强中效果更好。
수착계산궤시각연구적불단진보화도상초분변솔기술적불단완선，인검검측도상적분변솔증강기술이연성위인문관주적열점。본문기우인검검측도상분변솔증강적연구현상화전통Retinex산법적결함문제，인입궤하포락원리화다분변솔산법，제출료침대인검검측도상분변솔증강적개진Retinex산법。해산법재수입도상상직접구득일층궤하포락작위조도도상적고계，계이구득경물반사계수，병차채용다분변솔산법중적도상금자탑원리진행구해전국성적우화。실험결과표명，본문제출적개진산법교전통Retinex산법，재인검검측도상적분변솔증강중효과경호。
With the growing progress of computer vision research and the gradual improvement of image’s super-resolution technology, the resolution enhancement technique of face image detection has focused our attention. Based on the present research situation of face detection’s resolution enhancement and defects of traditional Retinex algorithm, the paper introduced geometric envelope principle and multiresolution algorithm, proposed the improved Retinex algorithm in view of face detection image resolution enhancement. The algorithm directly get the estimation based on one layer as the illumination image in the input image, and then get scenery reflection coefficient, and using the pyramid principle of the multiresolution algorithm to solve global optimization.The experimental results show that compared with the traditional Retinex algorithm, the proposed algorithm shows a better effect in the enhancement of face detection image resolution.