In this paper, a new approach for texture classification called wavelet domain majority coupled binary pattern is proposed. Here, the single-level wavelet transform is applied which decomposes the image, resulting in wavelet coefficients. The wavelet coefficients present in all the four sub-bands are taken for further processing. The relationship of wavelet coefficients present at distances one, two and three is utilized. The average wavelet coefficients present at various distances are compared with the center wavelet coefficient of the local region, resulting in binary value. For each distance, eight bit binary pattern is generated. Altogether, three distances yield three eight bit binary pattern. Then, the rule of majority is applied to the three eight bit binary pattern and results in generation of proposed label. The proposed labels together contribute for the construction of histogram. Finally, the distance measure is used to identify the similarity between query and database images. Experimental results show that the proposed method achieves the average retrieval rate of 88.92% on Brodatz, 93.95% on Outex and 90.53% on Virus databases. This shows that the proposed method achieves good performance and outperforms other existing methods.

本文提出了一种新的纹理分类方法，即小波域多数耦合二进制模式。在此，应用单级小波变换来分解图像，从而产生小波系数。将所有四个子带中都存在的小波系数用于进一步处理。利用存在于距离1、2和3处的小波系数的关系。将在不同距离处出现的平均小波系数与局部区域的中心小波系数进行比较，得出二进制值。对于每个距离，将生成八位二进制模式。总而言之，三个距离产生三个八位二进制模式。然后，将多数规则应用于三个八位二进制模式，并生成建议的标签。提议的标签共同为直方图的构建做出了贡献。最后，距离度量用于识别查询图像与数据库图像之间的相似性。实验结果表明，该方法在Brodatz上的平均检索率为88.92％，在Outex上的平均检索率为93.95％，在Virus数据库上的平均检索率为90.53％。这表明所提出的方法具有良好的性能，并且优于其他现有方法。