In this paper, an initial work of a research is discussed which is to teach young autistic children recognizing human facial expression with the help of computer vision and image processing. This paper mostly discusses the initial work of facial expression recognition using a deep convolutional neural network. The Kaggle's FER2013 dataset has been used to train and experiment with a deep convolutional neural network model. Once a satisfactory result is achieved, the dataset is modified with pictures of four different lighting conditions and each of these datasets is again trained with the same model. This is necessary for the end goal of the research which is to recognize facial expression in any possible environment. Finally, the comparison between results with different datasets is discussed and future work of the project is outlined.

PurposeDeep learning (DL) models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays (CXRs). Recently available public CXR datasets include high resolution images, but state-of-the-art models are trained on reduced size images due to limitations on graphics processing unit memory and training time. As computing hardware continues to advance, it has become feasible to train deep convolutional neural networks on high-resolution images without sacrificing detail by downscaling. This study examines the effect of increased resolution on CXR classification performance.ApproachWe used the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution CXR images for this study. We applied image downscaling from native resolution to 2048 × 2048 pixels, 1024 × 1024 pixels, 512 × 512 pixels, and 256 × 256 pixels and then we used the DenseNet121 and EfficientNet-B4 DL models to evaluate clinical task performance using these four downscaled image resolutions.ResultsWe find that while some clinical findings are more reliably labeled using high resolutions, many other findings are actually labeled better using downscaled inputs. We qualitatively verify that tasks requiring a large receptive field are better suited to downscaled low resolution input images, by inspecting effective receptive fields and class activation maps of trained models. Finally, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, indicating that diverse information is extracted across resolutions.ConclusionsThis study suggests that instead of focusing solely on the finest image resolutions, multi-scale features should be emphasized for information extraction from high-resolution CXRs.

In this paper, continued work of a research project is discussed whose end goal is to build a mobile device application that can teach children with ASD (Autism Spectrum Disorder) to recognize human facial expressions utilizing computer vision and image processing. This paper discusses the intermediate work of a facial expression recognition approach using a deep convolutional neural network (DCNN) utilizing images from different angles. The Karolinska Directed Emotional Faces (KDEF) dataset has been used to train and test with the DCNN model. This dataset contains images from five different angles. Results of this paper will contribute to the end goal of the research which is to recognize facial expression from any angle of viewpoint. Finally, the result obtained is discussed and future work of the project is outlined.

In this paper, continued work of a research project is discussed which achieved the end goal of the project - to build a mobile device application that can teach children with Autism Spectrum Disorder (ASD) to recognize human facial expressions utilizing computer vision and image processing. Universally, there are seven facial expressions categories: angry, disgust, happy, sad, fear, surprise, and neutral. To recognize all these facial expressions and to predict the current mood of a person is a difficult task for a child. A child with ASD, this problem presents itself in a more sophisticated manner due to the nature of the disorder. The main goal of this research was to develop a deep Convolutional Neural Network (DCNN) for facial expression recognition, which can help young children with ASD to recognize facial expressions, using mobile devices. The Kaggle's FER2013 and Karolinska Directed Emotional Faces (KDEF) dataset have been used to train and test with the DCNN model, which can classify facial expressions from different viewpoints and in different lighting contrasts. An 86.44% accuracy was achieved with good generalizability for the DCNN model. The results show an improvement of the DCNN accuracy in dealing with lighting contrast changes, and the implementation of image processing before performing the facial expression classification. As a byproduct of this research project, an app suitable for the iOS platform was developed for running both the DCNN model and image processing algorithm. The app can be used by speech-language pathologies, teacher, care-takers, and parents as a technological tool when working with children with ASD.

Deep learning models have received much attention lately for their ability to achieve expert-level performance on the accurate automated analysis of chest X-rays. Although publicly available chest X-ray datasets include high resolution images, most models are trained on reduced size images due to limitations on GPU memory and training time. As compute capability continues to advance, it will become feasible to train large convolutional neural networks on high-resolution images. To verify that this will lead to increased performance, we perform a systematic evaluation to measure the effect of input chest X-ray image resolution on accuracy. This study is based on the publicly available MIMIC-CXR-JPG dataset, comprising 377,110 high resolution chest X-ray images, and provided with 14 labels to the corresponding free-text radiology reports. Our original hypothesis that increased resolution would lead to higher accuracy held true for some but not all of the tasks. We find, interestingly, that tasks that require a large receptive field are better suited to downscaled input images, and we verify this qualitatively by inspecting effective receptive fields and class activation maps of trained models. Finally, we show that stacking an ensemble across resolutions outperforms each individual learner at all input resolutions while providing interpretable scale weights, suggesting that multi-scale features are crucially important to information extraction from high-resolution chest X-rays.