Retinal Optical Coherence Tomography (OCT) images, a non-invasive imaging technique, have become a standard retinal disease detection tool. Due to disease, there are morphological and textural changes in the layers of...Retinal Optical Coherence Tomography (OCT) images, a non-invasive imaging technique, have become a standard retinal disease detection tool. Due to disease, there are morphological and textural changes in the layers of the retina. Classifying OCT images is challenging, as the morphological manifestations of different diseases may be similar. The OCT images capture the reflectivity characteristics of the retinal tissues. Retinal diseases change the reflectivity property of retinal tissues, resulting in texture variations in OCT images. We propose a hybrid approach to OCT image classification in which the Convolution Neural Network (CNN) model is trained using Multiple Neighborhood Local Ternary Pattern (MNLTP) texture descriptors of the OCT images dataset for a robust disease prediction system. Parallel deep CNN (PDCNN) is proposed to improve feature representation and generalizability. The MNLTP-PDCNN model is tested on two publicly available datasets. The parameter values Accuracy, Precision, Recall, and F1-Score are calculated. The best accuracy obtained specifying the model’s overall performance is 93.98% and 99% for the NEH and OCT2017 datasets, respectively. With the proposed architecture, comparable performance is obtained with a subset of the original OCT2017 data set and a comparatively smaller number of trainable parameters (1.6 million, 1.8 million, and 2.3 million for a single CNN branch, two parallel CNN branches, and three parallel network branches, respectively), compared to off-the-shelf CNN models. Hence, the proposed approach is suitable for real-time OCT image classification systems with fast training of the CNN model and reduced memory requirement for computations.展开更多
Because texture images cannot be directly processed by the gray level information of individual pixel,we propose a new texture descriptor which reflects the intensity distribution of the patch centered at each pixel.T...Because texture images cannot be directly processed by the gray level information of individual pixel,we propose a new texture descriptor which reflects the intensity distribution of the patch centered at each pixel.Then the general multiphase image segmentation model of Potts model is extended for texture segmentation by adding the region information of the texture descriptor.A fast numerical scheme based on the split Bregman method is designed to speed up the computational process.The algorithm is efficient,and both the texture descriptor and the characteristic functions can be implemented easily.Experiments using synthetic texture images,real natural scene images and synthetic aperture radar images are presented to give qualitative comparisons between our method and other state-of-the-art techniques.The results show that our method can accurately segment object regions and is competitive compared with other methods especially in segmenting natural images.展开更多
Texture synthesis is widely used for modeling the appearance of virtual objects. However, traditional texture synthesis techniques eInphasize creation of optimal target textures, and pay insufficient attention to choi...Texture synthesis is widely used for modeling the appearance of virtual objects. However, traditional texture synthesis techniques eInphasize creation of optimal target textures, and pay insufficient attention to choice of suitable input texture exemplars. Currently, of taining texture exemplars from natural images is a labor intensive task for the artists, requiring careful photography and significant post- processing. In this paper, we present an automatic texture exemplar extraction method based on global and local textureness measures. To improve the efficiency of dominant texture identification, we first perform Poisson disk sampling to randomly and uniformly erop patches from a natural image. For global textureness assessment, we use a GIST descriptor to distinguish textured t）atches from non-textured patches, in conjunction with SVM prediction. To identify real texture, exemplars consisting solely of the dominant texture, we further measure the local textureness of a patch by extracting and matching the local structure （using t）inary Gabor pattern （BGP）） and dominant color features （using color histograms） between a patch and its sub-regions. Finally, we obtain optimal texture exemplars by scoring and ranking extracted patches using these global and local textureness measures We evaluate our method on a variety of images with different kinds of textures. A convincing visual comparison with textures mauually selected by an artist and a statistical study demonstrate its effectiveness.展开更多
基金Deanship of Research and Graduate Studies at King Khalid University funded this work through Large Research Project under grant number RGP2/54/45.
文摘Retinal Optical Coherence Tomography (OCT) images, a non-invasive imaging technique, have become a standard retinal disease detection tool. Due to disease, there are morphological and textural changes in the layers of the retina. Classifying OCT images is challenging, as the morphological manifestations of different diseases may be similar. The OCT images capture the reflectivity characteristics of the retinal tissues. Retinal diseases change the reflectivity property of retinal tissues, resulting in texture variations in OCT images. We propose a hybrid approach to OCT image classification in which the Convolution Neural Network (CNN) model is trained using Multiple Neighborhood Local Ternary Pattern (MNLTP) texture descriptors of the OCT images dataset for a robust disease prediction system. Parallel deep CNN (PDCNN) is proposed to improve feature representation and generalizability. The MNLTP-PDCNN model is tested on two publicly available datasets. The parameter values Accuracy, Precision, Recall, and F1-Score are calculated. The best accuracy obtained specifying the model’s overall performance is 93.98% and 99% for the NEH and OCT2017 datasets, respectively. With the proposed architecture, comparable performance is obtained with a subset of the original OCT2017 data set and a comparatively smaller number of trainable parameters (1.6 million, 1.8 million, and 2.3 million for a single CNN branch, two parallel CNN branches, and three parallel network branches, respectively), compared to off-the-shelf CNN models. Hence, the proposed approach is suitable for real-time OCT image classification systems with fast training of the CNN model and reduced memory requirement for computations.
基金supported by the National Natural Science Foundation of China(No.61170106)
文摘Because texture images cannot be directly processed by the gray level information of individual pixel,we propose a new texture descriptor which reflects the intensity distribution of the patch centered at each pixel.Then the general multiphase image segmentation model of Potts model is extended for texture segmentation by adding the region information of the texture descriptor.A fast numerical scheme based on the split Bregman method is designed to speed up the computational process.The algorithm is efficient,and both the texture descriptor and the characteristic functions can be implemented easily.Experiments using synthetic texture images,real natural scene images and synthetic aperture radar images are presented to give qualitative comparisons between our method and other state-of-the-art techniques.The results show that our method can accurately segment object regions and is competitive compared with other methods especially in segmenting natural images.
基金supported in part by grants from the National Natural Science Foundation of China(Nos.61303101 and 61572328)the Shenzhen Research Foundation for Basic Research,China(Nos.JCYJ20150324140036846,JCYJ20170302153551588,CXZZ20140902160818443,CXZZ20140902102350474,CXZZ20150813151056544,JCYJ20150630105452814,JCYJ20160331114551175,and JCYJ20160608173051207)the Startup Research Fund of Shenzhen University(No.2013-827-000009)
文摘Texture synthesis is widely used for modeling the appearance of virtual objects. However, traditional texture synthesis techniques eInphasize creation of optimal target textures, and pay insufficient attention to choice of suitable input texture exemplars. Currently, of taining texture exemplars from natural images is a labor intensive task for the artists, requiring careful photography and significant post- processing. In this paper, we present an automatic texture exemplar extraction method based on global and local textureness measures. To improve the efficiency of dominant texture identification, we first perform Poisson disk sampling to randomly and uniformly erop patches from a natural image. For global textureness assessment, we use a GIST descriptor to distinguish textured t）atches from non-textured patches, in conjunction with SVM prediction. To identify real texture, exemplars consisting solely of the dominant texture, we further measure the local textureness of a patch by extracting and matching the local structure （using t）inary Gabor pattern （BGP）） and dominant color features （using color histograms） between a patch and its sub-regions. Finally, we obtain optimal texture exemplars by scoring and ranking extracted patches using these global and local textureness measures We evaluate our method on a variety of images with different kinds of textures. A convincing visual comparison with textures mauually selected by an artist and a statistical study demonstrate its effectiveness.
