双密度双树复小波RH模型纹理图像检索

Texture image retrieval using double density dual tree complex wavelet RH model

针对利用光滑概率密度函数提取小波系数直方图及系数模直方图方法存在参数估计复杂,难以充分有效地提取纹理特征的不足。提出了双密度双树复小波RH模型的纹理图像检索方法,通过分析双密度双树复小波原理,RH模型与非均匀量化器的内在关系,将RH模型推广为提取双密度双树复小波变换系数及系数模直方图特征。该方法结合了RH模型及双密度双树复小波的优点。实验表明,研究方法与利用概率密度函数提取提取直方图特征的方法相比检索率提高了2～9%;推广RH模型提取双密度双树复小波系数模特征的方法获得了75.66%的最高检索率。

For using smooth probability density function to retrieve wavelet coefficient histogram and coefficient module histogram, parameter estimation is complicated, which results in hard to retrieve the texture features effectively. A texture image retrieval method using double density dual tree complex wavelet Refined Histogram（RH） model is proposed. By analyzing the principle of double density dual tree complex wavelet transform （DD-DT CWT） and the inherent relationship between the nonuniform quantizer and RH model, the RH model is extended to retrieve the DD-DT CWT coefficient and the coefficient histogram feature. The RH is used to model the magnitude of the DD-DT CWT. The RH parameters for all magnitude of complex coefficients forms the signature of an image. Image similarity measurement is accomplished by using the Kullback-Leibler divergences . The proposed method combines the advantages of the RH model and the shift-invariant DD-DT CWT. The experiment results show that the proposed methods yields higher retrieval rate than using the General Gaussian Density（GGD） model to fit with the real part or imaginary part of coefficients, and is better than using the Gamma PDF to fit with the magnitude of coefficients.