Abstract:

Compressed-domain image analysis and process have been an active area in the fields of multimedia research. A fast JPEG image edge detection method based on DCT compressed domain was presented, which computes the edge points directly from none-zero DCT coefficients without full decompression, so that a lot of computational cost was saved. In addition, edge image with different accuracy to meet different demands could be obtained progressively by this method. Compared with the traditional pixel domain based edge detection algorithms such as Laplacian Operator, Canny Operator, Sobel Operator and so on, the proposed method calculated the pixel gradient directly from some important DCT coefficients other than original pixel values. Further more, some approaches to reduce the computational cost were used in this paper. The method could be used in the long- range object recognition or Web-based JPEG image retrieval.

Key words: compressed domain edge detection, DCT, JPEG

摘要：

压缩域的图像分析处理技术已成为多媒体研究领域的一个热点。文中给出了DCT压缩域图像边缘的快速检测方法。该方法直接利用DCT非零系数计算图像边缘点,不需要完全解压缩,与传统象素域边缘检测方法相比,大大降低了计算复杂度,并且能根据需要提取不同精度的边缘图像。该方法在远程目标识别或基于边缘的Web图像检索等方面将能满足一定的实时性要求,具有较好的实用价值。针对JPEG图像,给出了边缘提取的实验结果,并与传统的象素域边缘检测方法进行了比较。

关键词: 压缩域边缘检测, DCT, JPEG