The multi-source image fusion has been a hot topic during recent years because of its higher segmentation accuracy rate. However, the traditional multi-source image fusion methods could not obtain better contrast and more details of the fused image. To better detect the pig-body feature, a novel infrared and visible image fusion method for pig-body segmentation and temperature detection is proposed in non-subsampled contourlet transform (NSCT) domain, named as NSCT-GF. Firstly, the visible and infrared images were decomposed into a series of multi-scale and multi-directional sub-bands using NSCT. Then, to better represent the fine-scale of texture information, the Gabor energy map was extracted by Gabor filter with even-symmetry, and the low-frequency coefficients were fused by the maximum of Ordinal encoding. Then, to preserve the more coarse-scale and edge detail information, Gabor filter with odd-symmetry was employed to fuse high-frequency NSCT sub-bands and the fused coefficients were reconstructed into a final fusion image by inverse NSCT. Next, the pig-body shape was obtained by Ostu automatic threshold segmentation and optimized by morphological processing. Finally, the pig-body temperature was extracted based on shape segmentation. Experimental results showed that the proposed segmentation method was capable of achieving 1.84–3.89% higher average segmentation accuracy rate than the prevailing conventional methods and also improved efficiency in terms of time consumption. It lays a foundation for accurately measuring the temperature of pig-body.

中文翻译：

---

一种基于Gabor特征的多源猪体多特征融合新方法

多源图像融合因其较高的分割准确率而成为近年来的热门话题。然而，传统的多源图像融合方法无法获得更好的对比度和融合图像的更多细节。为了更好地检测猪体特征，在非下采样轮廓波变换（NSCT）域中提出了一种用于猪体分割和温度检测的红外和可见光图像融合新方法，命名为NSCT-GF。首先，使用NSCT将可见光和红外图像分解为一系列多尺度多方向的子带。然后，为了更好地表示纹理信息的精细尺度，采用偶对称的Gabor滤波器提取Gabor能量图，并通过Ordinal编码的最大值融合低频系数。然后，为了保留更粗的尺度和边缘细节信息，采用奇对称Gabor滤波器融合高频NSCT子带，融合系数通过逆NSCT重建为最终融合图像。接下来，通过Ostu自动阈值分割获得猪体形状，并通过形态学处理进行优化。最后，基于形状分割提取猪体温。实验结果表明，所提出的分割方法能够实现比流行的传统方法高1.84-3.89%的平均分割准确率，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。采用奇对称 Gabor 滤波器对高频 NSCT 子带进行融合，融合后的系数通过逆 NSCT 重建为最终融合图像。接下来，通过Ostu自动阈值分割获得猪体形状，并通过形态学处理进行优化。最后，基于形状分割提取猪体温。实验结果表明，所提出的分割方法能够实现比流行的传统方法高1.84-3.89%的平均分割准确率，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。采用奇对称 Gabor 滤波器对高频 NSCT 子带进行融合，融合后的系数通过逆 NSCT 重建为最终融合图像。接下来，通过Ostu自动阈值分割获得猪体形状，并通过形态学处理进行优化。最后，基于形状分割提取猪体温。实验结果表明，所提出的分割方法能够实现比流行的传统方法高1.84-3.89%的平均分割准确率，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。通过Ostu自动阈值分割获得猪体形状，并通过形态学处理进行优化。最后，基于形状分割提取猪体温。实验结果表明，所提出的分割方法能够实现比流行的传统方法高1.84-3.89%的平均分割准确率，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。通过Ostu自动阈值分割获得猪体形状，并通过形态学处理进行优化。最后，基于形状分割提取猪体温。实验结果表明，所提出的分割方法能够实现比流行的传统方法高1.84-3.89%的平均分割准确率，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。平均分割准确率比流行的传统方法高89%，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。平均分割准确率比流行的传统方法高89%，并且在时间消耗方面也提高了效率。为准确测量猪体温度奠定基础。