The Experts below are selected from a list of 360 Experts worldwide ranked by ideXlab platform
Abir Bhowmick - One of the best experts on this subject based on the ideXlab platform.
-
script identification in natural scene image and video frames using an attention based convolutional lstm network
Pattern Recognition, 2019Co-Authors: Ankan Kumar Bhunia, Aishik Konwer, Ayan Kumar Bhunia, Abir BhowmickAbstract:Abstract Script identification plays a significant role in analysing documents and videos. In this paper, we focus on the problem of script identification in scene text images and video scripts. Because of low image quality, complex background and similar layout of characters shared by some scripts like Greek, Latin, etc., text recognition in those cases become challenging. In this paper, we propose a novel method that involves extraction of local and global features using CNN-LSTM framework and weighting them dynamically for script identification. First, we convert the images into patches and feed them into a CNN-LSTM framework. Attention-based patch weights are calculated applying softmax layer after LSTM. Next, we do patch-wise multiplication of these weights with corresponding CNN to yield local features. Global features are also extracted from last cell state of LSTM. We employ a fusion technique which dynamically weights the local and global features for an individual patch. Experiments have been done in four public script identification datasets: SIW-13, CVSI2015, ICDAR-17 and MLe2e. The proposed framework achieves superior results in comparison to conventional methods.
Marcus Taft - One of the best experts on this subject based on the ideXlab platform.
-
subsyllabic structure reflected in letter confusability effects in korean word recognition
Psychonomic Bulletin & Review, 2011Co-Authors: Marcus TaftAbstract:Korean subsyllabic structure was investigated by observing the pattern of responses arising from letter transpositions within a syllable in the Hangul script. Experiment 1 revealed no confusions when the onset and coda of one syllable of a disyllabic word were transposed. This was also the case in Experiment 2, where the transposition took place within a syllable having a coda composed of two consonants. However, confusion with the baseword was observed when the two consonants within the complex coda were themselves transposed. These results are taken to support the view that Hangul (and possibly all concatenating orthographic scripts) is processed based on its onset, vowel, and coda structure.
-
subsyllabic structure reflected in letter confusability effects in korean word recognition
Psychonomic Bulletin & Review, 2011Co-Authors: Marcus TaftAbstract:Korean subsyllabic structure was investigated by observing the pattern of responses arising from letter transpositions within a syllable in the Hangul script. Experiment 1 revealed no confusions when the onset and coda of one syllable of a disyllabic word were transposed. This was also the case in Experiment 2, where the transposition took place within a syllable having a coda composed of two consonants. However, confusion with the baseword was observed when the two consonants within the complex coda were themselves transposed. These results are taken to support the view that Hangul (and possibly all concatenating orthographic scripts) is processed based on its onset, vowel, and coda structure.
Ankan Kumar Bhunia - One of the best experts on this subject based on the ideXlab platform.
-
script identification in natural scene image and video frames using an attention based convolutional lstm network
Pattern Recognition, 2019Co-Authors: Ankan Kumar Bhunia, Aishik Konwer, Ayan Kumar Bhunia, Abir BhowmickAbstract:Abstract Script identification plays a significant role in analysing documents and videos. In this paper, we focus on the problem of script identification in scene text images and video scripts. Because of low image quality, complex background and similar layout of characters shared by some scripts like Greek, Latin, etc., text recognition in those cases become challenging. In this paper, we propose a novel method that involves extraction of local and global features using CNN-LSTM framework and weighting them dynamically for script identification. First, we convert the images into patches and feed them into a CNN-LSTM framework. Attention-based patch weights are calculated applying softmax layer after LSTM. Next, we do patch-wise multiplication of these weights with corresponding CNN to yield local features. Global features are also extracted from last cell state of LSTM. We employ a fusion technique which dynamically weights the local and global features for an individual patch. Experiments have been done in four public script identification datasets: SIW-13, CVSI2015, ICDAR-17 and MLe2e. The proposed framework achieves superior results in comparison to conventional methods.
Herta Flor - One of the best experts on this subject based on the ideXlab platform.
-
dissociation proneness and pain hyposensitivity in current and remitted borderline personality disorder
European Journal of Pain, 2020Co-Authors: Boo Young Chung, Saskia Hensel, Ilinca Schmidinger, Robin Bekraterbodmann, Herta FlorAbstract:BACKGROUND: Stress-related dissociation has been shown to negatively co-vary with pain perception in current borderline personality disorder (cBPD). While remission of the disorder (rBPD) is associated with normalized pain perception, it remains unclear whether dissociation proneness is still enhanced in this group and how this feature interacts with pain sensitivity. METHODS: Twenty-five cBPD patients, 20 rBPD patients and 24 healthy controls (HC) participated in an experiment using the script-driven imagery approach. We presented a personalized stressful and neutral narrative. After listening to the scripts, dissociation and heat pain thresholds (HPT) were assessed. RESULTS: Compared to HC, cBPD patients showed enhanced dissociation and exhibited significantly enhanced HPT in the neutral condition, while rBPD participants were in between. After listening to the stress script, both clinical groups exhibited enhanced dissociation scores. Current BPD participants responded with significantly higher HPT, while rBPD only showed a trend in the same direction. However, both BPD groups showed significantly increased HPT compared to the HC in the stress condition, but did not differ from each other. Dissociation proneness correlated significantly positively with pain hyposensitivity only in cBPD. CONCLUSION: Dissociation proneness is enhanced in both BPD groups. This feature is clearly positively related to pain hyposensitivity in cBPD, but not in rBPD. However, the data indicate that stress causes the pain perception in rBPD to drift away from that obtained in HC. These results highlight the volatile state of BPD remission and might have important implications for the care of BPD patients in the remitted stage. SIGNIFICANCE: Both current (cBPD) and remitted borderline personality disorder (rBPD) patients show enhanced proneness to dissociation. This feature is significantly linked with pain hyposensitivity in cBPD in a paradigm that induces stress using a script-driven imagery approach, whereas this connection cannot be observed in rBPD. However, in the stress compared to the neutral condition, rBPD participants also show pain hyposensitivity compared to healthy controls. This study provides new insights into the pain processing mechanisms of BPD and its remission.
Ayan Kumar Bhunia - One of the best experts on this subject based on the ideXlab platform.
-
script identification in natural scene image and video frames using an attention based convolutional lstm network
Pattern Recognition, 2019Co-Authors: Ankan Kumar Bhunia, Aishik Konwer, Ayan Kumar Bhunia, Abir BhowmickAbstract:Abstract Script identification plays a significant role in analysing documents and videos. In this paper, we focus on the problem of script identification in scene text images and video scripts. Because of low image quality, complex background and similar layout of characters shared by some scripts like Greek, Latin, etc., text recognition in those cases become challenging. In this paper, we propose a novel method that involves extraction of local and global features using CNN-LSTM framework and weighting them dynamically for script identification. First, we convert the images into patches and feed them into a CNN-LSTM framework. Attention-based patch weights are calculated applying softmax layer after LSTM. Next, we do patch-wise multiplication of these weights with corresponding CNN to yield local features. Global features are also extracted from last cell state of LSTM. We employ a fusion technique which dynamically weights the local and global features for an individual patch. Experiments have been done in four public script identification datasets: SIW-13, CVSI2015, ICDAR-17 and MLe2e. The proposed framework achieves superior results in comparison to conventional methods.