Existing popular descriptors for facial expression recognition often suffer from inconsistent feature description, experiencing poor accuracies. We present a new local descriptor, local directional-structural pattern (LDSP), in this work to address this issue. Unlike the existing local descriptors using only the texture or edge information to represent the local structure of a pixel, the proposed LDSP utilizes the positional relationship of the top edge responses of the target pixel to extract more detailed structural information of the local texture. We further exploit such information to characterize expression-affiliated crucial textures while discarding the random noisy patterns. Moreover, we introduce a globally adaptive thresholding strategy to exclude futile flat patterns. Hence, LDSP offers a stable description of facial expressions with the explicit representation of the expression-affiliated features along with the exclusion of random futile textures. We visualize the efficacy of the proposed method in three folds. First, the LDSP descriptor possesses a moderate code-length owing to the exclusion of the futile patterns, yielding less computation time than other edge descriptors. Second, for person-independent expression recognition in benchmark datasets, LDSP demonstrates higher accuracy than existing descriptors and other state-of-the-art methods. Third, LDSP shows better performance than other descriptors against noise and low resolution, exhibiting its robustness under such uneven conditions.
In recent years, researchers have been working to interpret the insights of deep neural networks in the pursuit of overcoming their opaqueness and so-called 'black-box' tag. In this work, we present a new posthoc visual interpretation technique that finds out discriminative image regions contributing highly towards networks' prediction. We select most discernible set of neurons per layer and engineer the forward pass operation to gradually reach most discriminative image locations. While searching for discernible neurons, existing approaches either end up with low-resolution visualization maps, or suffer lack of neuron discriminativity in the way. Moreover, some methods concentrate only on current layer information overlooking meaningful information from adjacent layers, limiting the overall scope of selection. We address these issues by exploring the layer-to-layer connected structure of a neuron and obtaining contributions from its current layer along with its adjacent layers, e.g., succeeding and preceding layer. We introduce a score function where such contributions are assembled with appropriate priorities. Layerwise discernible neurons are then selected based on top scores, ensuring a reliable and credible selection. We validate our proposal through objective and subjective evaluations by examining its performance in terms of models' faithfulness and human-trust, where we visualize its efficacy over other existing methods. We also perform sets of sanity check experiments on our method to show its overall reliability as a visualization map.
Human age recognition from face image relies highly on a reasonable aging description. Considering the disparate and complex face-aging variation of each person, aging description needs to be defined carefully with detailed local information. However, aging description relies highly on the appropriate definition of different aging-affiliated textures. Wrinkles are considered as the most discernible textures in this regard owing to their significant visual appearance in human aging. Most of the existing image-descriptors, however, fail short to preserve diverse variations of wrinkles, such as a) characterizing stronger and smoother wrinkles, appropriately, b) distinguishing wrinkles from nonwrinkle patterns, and c) characterizing the proper texturestructures of the pixels belonging to the same wrinkle. In this paper, we address these issues by presenting a new local descriptor, Local Edge-Prototypic Pattern (LEPP) with the notion that LEPP preserves different variations of wrinkle-patterns appropriately in representing the aging description. In the coding, LEPP sets prototypic restrictions for each neighboring pixel using their relation with center pixel when they belong to an inlying-edge, and utilize such restrictions, afterwards, to prioritize specific neighbors showing significant edge-signature. This strategy appropriately encodes the inlying edge structure of aging-affiliated textures and simultaneously, avoids featureless texture. We visualize the stability of LEPP in terms of its robustness under noise. Our experiments show that LEPP preserves discernible aging variations yielding better accuracies than the state-of-theart methods in popular age-group datasets.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.