基于自适应Gabor小波包的超声心动图像纹理特征提取

在基于纹理的医学图像检索中，由于提取的图像特征向量过长，大大影响检索速度．为此，提出一种基于Gabor小波包的自适应纹理提取方法：以信号能量集中度作为自适应选取小波包的优化参数，建立反映图像内容的纹理特征向量以及空间信息结构码．这种方法有效降低了纹理特征向量的维数，显著提高了图像分析的效率．在基于内容的图像检索实验中，以两类典型的超声心动图像为研究对象，结果表明该方法能够达到95．1％的平均识别率．

Texture Feature Extraction of Ultrasound Cardiogram Based on Adaptive Gabor Wavelet Packet

In the texture-based retrievals of physic image, the speed of the process is largely slowed down because of the large size of the texture feature vector extracted from the image. To resolve this problem, a feature extraction method is proposed based on adaptive Gabor wavelet packet. Energy concentration is used as the prime index to optimize wavelet base set and the space of wavelet packet. Then a feature vector and a space information structure code reflecting the texture of the image are established. While effectively reducing the dimensions of the vector, this method remarkably increases the efficiency of analysis on ultrasound cardiogram. In the retrievals of physic image based on the content of ultrasound cardiogram, two kinds of typical ultrasound cardiogram are used as research objects. An average identification rate of 95.1% is achieved.