GPU并行实现的基于多方向线状梯度调节预测器边缘检测算法

针对无损压缩编码中梯度调节预测器(GAP)模板的方向固定、单一的问题，根据实际边缘具有线状变化增量相同的特征，提出多方向线状梯度调节预测器(MLGAP)模板。首先从图像中心向四周划分四个子图像，应用图形处理器(GPU)并行技术，在每个子图像中采用MLGAP模板计算预测值；然后利用错误反馈信息构建预测误差图像；再通过大津(OTSU)算法计算阈值；分类误差图像边缘；最后用Hilditch算法细化边缘。实验结果表明，图像边缘检测定位精确，噪声少，细节丰富，而且GPU并行技术加速了图像处理。

GPU parallel implementation of edge-detection algorithm based on multidirectional linear gradient adjusted predictor

Concerning the fixed direction and monotony of lossless image compression template of Gradient Adjusted Predictor (GAP), according to the characteristic of the actual edge with the same linear increments, this paper proposed Multidirectional Linear Gradient Adjusted Predictor (MLGAP) template. Firstly the image was cut into four sub-images from the center to the periphery. With the application of Graphics Processor Unit (GPU) parallel technology operation, the predictive value was calculated by MLGAP template in each sub-image, and then error feedback was used to construct prediction error image. The threshold was calculated by OSTU algorithm, and error image edge was classified. At last the edge was thinned by Hilditch algorithm. The simulation results show that using the proposed method can get clear, complete and precise edges. In addition, the GPU parallel technology accelerates the image processing.