Sihyoung Lee

Multiple ROI selection based focal liver lesion classification in ultrasound images

Ultrasound imaging is one of the most widely used imaging modality for the purpose of visualizing the human soft tissues. Especially, liver imaging application is of great importance in the areas of diagnostic ultrasound. In ultrasound liver image, the classification of lesions depends heavily on the characteristics of the lesions including internal echo, morphology, edge, echogenicity, and posterior echo enhancement. These characteristics are differently observed according to ROI selection methods that may indeed significantly impact the classification performances. Currently developed ROI selection methods have limitation for guaranteeing robust classification performance for focal liver lesions, mainly due to the inherent difficulties that represent all ultrasonic appearances of characteristics of lesion. In order to obtain better and more stable classification performances, we propose a new and novel approach, so-called multiple-ROI based focal liver lesion classification. The proposed approach properly combines the advantages of existing ROI selection methods to represent well various ultrasonic appearances of liver lesions including internal echo, morphology, edge, echogenicity, and posterior echo enhancement. To verify the effectiveness of the proposed ROI selection approach, extensive and comparative experiments have been performed using a total of 150 ultrasound images. Each ultrasound image contains one corresponding focal liver lesion so that a total of 150 focal liver lesions is used, comprising of 50 cysts, 50 hemangiomas, and 50 malignancies. Experimental results show that the proposed multiple-ROI-based approach can achieve the enhanced and stable classification performance regardless of features being used. In addition, our proposed method outperforms other existing classification methods designed for focal liver lesion classification. Especially, the proposed approach attains classification accuracy of up to 80% over well-known challenging task of classifying the hemangiomas and malignancies.

MAP-based image tag recommendation using a visual folksonomy

Descriptive tags are needed to enable efficient and effective search in vast collections of images. Tag recommendation represents a trade-off between automatic image annotation techniques and manual tagging. In this letter, we formulate image tag recommendation as a maximum a posteriori (MAP) problem, making use of a visual folksonomy. A folksonomy can be seen as a collaboratively created set of metadata for informal social classification. Our experimental results show that the use of a visual folksonomy for image tag recommendation has two significant benefits, compared to a conventional approach using a limited concept vocabulary. First, our tag recommendation technique can make use of an unrestricted and rich concept vocabulary. Second, our approach is able to recommend a higher number of correct tags.

Tag refinement in an image folksonomy using visual similarity and tag co-occurrence statistics

Noisy tag assignments lower the effectiveness of multimedia applications that rely on the availability of user-supplied tags for retrieving user-contributed images for further processing. This paper discusses a novel tag refinement technique that aims at differentiating noisy tag assignments from correct tag assignments. The correctness of tag assignments is determined through the combined use of visual similarity and tag co-occurrence statistics. To verify the effectiveness of our tag refinement technique, experiments were performed with user-contributed images retrieved from Flickr. For the image set used, the proposed tag refinement technique reduces the number of noisy tag assignments with 36% (benefit), while removing 10% of the correct tag assignments (cost). In addition, our tag refinement technique increases the effectiveness of tag recommendation for non-annotated images with 45% when using the P@5 metric and with 41% when using the NDCG metric.

Image tag refinement along the ‘what’ dimension using tag categorization and neighbor voting

Online sharing of images is increasingly becoming popular, resulting in the availability of vast collections of user-contributed images that have been annotated with usersupplied tags. However, user-supplied tags are often not related to the actual image content, affecting the performance of multimedia applications that rely on tag-based retrieval of user-contributed images. This paper proposes a modular approach towards tag refinement, taking into account the nature of tags. First, tags are automatically categorized in five categories using WordNet: ‘where’, ‘when’, ‘who’, ‘what’, and ‘how’. Next, as a start towards a full implementation of our modular tag refinement approach, we use neighbor voting to learn the relevance of tags along the ‘what’ dimension. Our experimental results show that the proposed tag refinement technique is able to successfully differentiate correct tags from noisy tags along the ‘what’ dimension. In addition, we demonstrate that the proposed tag refinement technique is able to improve the effectiveness of image tag recommendation for non-tagged images.

Improving image tag recommendation using favorite image context

Tag recommendation allows mitigating the amount of user effort needed to annotate images. Assuming that favorite images and their associated tags are indicative of the visual and topical interests of users, this paper proposes a personalized image tag recommendation technique that makes use of favorite image context. Specifically, to recommend tags for a newly uploaded image, we propose to take advantage of the tags assigned to favorite images of the user who uploaded the image, fusing tag statistics and visual similarity. Experimental results obtained for images and tags retrieved from Flickr compare the use of favorite image context to the use of personal and collective context for the purpose of tag recommendation, showing that the use of favorite image context is promising.

Semantic Photo Album Based on MPEG-4 Compatible Application Format

In this paper, we propose a promising photo album system that enables augmented use of digital photos over a wide range of multimedia devices and semantic consumption of photos as well. The proposed system contains an album engine to create metadata and a photo player to consume the photos. It operates based on a standardized application format to encode a group of photos and associated metadata. Experiments showed that the proposed system achieved reasonable albuming performance.

Towards data-driven estimation of image tag relevance using visually similar and dissimilar folksonomy images

Given that the presence of non-relevant tags in an image folksonomy hampers the effective organization and retrieval of images, this paper discusses a novel technique for estimating the relevance of user-supplied tags with respect to the content of a seed image. Specifically, this paper proposes to compute the relevance of image tags by making use of both visually similar and dissimilar images. That way, compared to tag relevance estimation only using visually similar images, the difference in tag relevance between tags relevant and tags irrelevant with respect to the content of a seed image can be increased at a limited increase in computational cost, thus making it more straightforward to distinguish between them. The latter is confirmed through experimentation with subsets of MIRFLICKR-25000 and MIRFLICKR-1M, showing that tag relevance estimation using both visually similar and dissimilar images allows achieving more effective image tag refinement and tag-based image retrieval than tag relevance estimation only using visually similar images.

Enhanced classification of focal hepatic lesions in ultrasound images using novel texture features

This paper discusses novel texture features that allow providing enhanced classification accuracy for focal hepatic lesions. The proposed texture features takes advantage of the rotation and scale invariant nature of Gabor wavelets, as well as the gray-level co-occurrence matrix (GLCM) for analyzing the spatial distribution of the pixel intensity in the lesion. To verify the effectiveness of the proposed texture features, experiments were performed with 150 ultrasound images containing 150 focal hepatic lesions, consisting of 50 cysts, 50 hemangiomas, and 50 malignancies. Experimental results show that the proposed texture features allow for an improved classification performance, compared to the use of other features.

Semantic concept detection for user-generated video content using a refined image folksonomy

The automatic detection of semantic concepts is a key technology for enabling efficient and effective video content management. Conventional techniques for semantic concept detection in video content still suffer from several interrelated issues: the semantic gap, the imbalanced data set problem, and a limited concept vocabulary size. In this paper, we propose to perform semantic concept detection for user-created video content using an image folksonomy in order to overcome the aforementioned problems. First, an image folksonomy contains a vast amount of user-contributed images. Second, a significant portion of these images has been manually annotated by users using a wide variety of tags. However, user-supplied annotations in an image folksonomy are often characterized by a high level of noise. Therefore, we also discuss a method that allows reducing the number of noisy tags in an image folksonomy. This tag refinement method makes use of tag co-occurrence statistics. To verify the effectiveness of the proposed video content annotation system, experiments were performed with user-created image and video content available on a number of social media applications. For the datasets used, video annotation with tag refinement has an average recall rate of 84% and an average precision of 75%, while video annotation without tag refinement shows an average recall rate of 78% and an average precision of 62%.