An image super-resolution processing by using Hidden Markov Tree model to express the wavelet coefficients of the probability of structure is proposed. According to the properties of multiscale wavelet coefficients, the persistency and the exponential decay, wavelet coefficients of the image are modeled as Hidden Markov Tree model. Wavelet-domain HMT model considers the statistical properties of wavelet coefficients. The image interpolation is formulated as a constrained optimization problem. The high-resolution interpolated image which is able to maintain the original image information in the rich high-frequency is obtained. Experiment results show that the algorithm improves the traditional sawtooth effect and smoothing effect caused by interpolation. The PSNR and visual effects of interpolated image are improved obviously.%为实现图像的超分辨率处理,提出利用HMT模型简洁地表示小波系数的概率结构.小波域HMT模型根据小波系数尺度之间的持续性和指数衰减性,将图像的小波系数建模为隐马尔可夫树模型.该模型考虑了小波系数间的统计相关性,把图像插值问题表述为一个约束优化问题,获得了能够保持原始图像丰富高频信息的高分辨率插值图像.试验结果表明,该算法在一定程度上改善了传统插值算法引起的锯齿效应和平滑效应,插值后的图像在峰值信噪比和视觉效果方面都有明显提高.
展开▼
