Multimodal Fusion Strategy Research Articles

Hi-Net: Hybrid-Fusion Network for Multi-Modal MR Image Synthesis.

Magnetic resonance imaging (MRI) is a widely used neuroimaging technique that can provide images of different contrasts (i.e., modalities). Fusing this multi-modal data has proven particularly effective for boosting model performance in many tasks. However, due to poor data quality and frequent patient dropout, collecting all modalities for every patient remains a challenge. Medical image synthesis has been proposed as an effective solution, where any missing modalities are synthesized from the existing ones. In this paper, we propose a novel Hybrid-fusion Network (Hi-Net) for multi-modal MR image synthesis, which learns a mapping from multi-modal source images (i.e., existing modalities) to target images (i.e., missing modalities). In our Hi-Net, a modality-specific network is utilized to learn representations for each individual modality, and a fusion network is employed to learn the common latent representation of multi-modal data. Then, a multi-modal synthesis network is designed to densely combine the latent representation with hierarchical features from each modality, acting as a generator to synthesize the target images. Moreover, a layer-wise multi-modal fusion strategy effectively exploits the correlations among multiple modalities, where a Mixed Fusion Block (MFB) is proposed to adaptively weight different fusion strategies. Extensive experiments demonstrate the proposed model outperforms other state-of-the-art medical image synthesis methods.

Study of Subjective and Objective Quality Assessment of Audio-Visual Signals.

The topics of visual and audio quality assessment (QA) have been widely researched for decades, yet nearly all of this prior work has focused only on single-mode visual or audio signals. However, visual signals rarely are presented without accompanying audio, including heavy-bandwidth video streaming applications. Moreover, the distortions that may separately (or conjointly) afflict the visual and audio signals collectively shape user-perceived quality of experience (QoE). This motivated us to conduct a subjective study of audio and video (A/V) quality, which we then used to compare and develop A/V quality measurement models and algorithms. The new LIVE-SJTU Audio and Video Quality Assessment (A/V-QA) Database includes 336 A/V sequences that were generated from 14 original source contents by applying 24 different A/V distortion combinations on them. We then conducted a subjective A/V quality perception study on the database towards attaining a better understanding of how humans perceive the overall combined quality of A/V signals. We also designed four different families of objective A/V quality prediction models, using a multimodal fusion strategy. The different types of A/V quality models differ in both the unimodal audio and video quality prediction models comprising the direct signal measurements and in the way that the two perceptual signal modes are combined. The objective models are built using both existing state-of-the-art audio and video quality prediction models and some new prediction models, as well as quality-predictive features delivered by a deep neural network. The methods of fusing audio and video quality predictions that are considered include simple product combinations as well as learned mappings. Using the new subjective A/V database as a tool, we validated and tested all of the objective A/V quality prediction models. We will make the database publicly available to facilitate further research.

How to Teach DNNs to Pay Attention to the Visual Modality in Speech Recognition

Audio-Visual Speech Recognition (AVSR) seeks to model, and thereby exploit, the dynamic relationship between a human voice and the corresponding mouth movements. A recently proposed multimodal fusion strategy, AV Align, based on state-of-the-art sequence to sequence neural networks, attempts to model this relationship by explicitly aligning the acoustic and visual representations of speech. This study investigates the inner workings of AV Align and visualises the audio-visual alignment patterns. Our experiments are performed on two of the largest publicly available AVSR datasets, TCD-TIMIT and LRS2. We find that AV Align learns to align acoustic and visual representations of speech at the frame level on TCD-TIMIT in a generally monotonic pattern. We also determine the cause of initially seeing no improvement over audio-only speech recognition on the more challenging LRS2. We propose a regularisation method which involves predicting lip-related Action Units from visual representations. Our regularisation method leads to better exploitation of the visual modality, with performance improvements between 7% and 30% depending on the noise level. Furthermore, we show that the alternative Watch, Listen, Attend, and Spell network is affected by the same problem as AV Align, and that our proposed approach can effectively help it learn visual representations. Our findings validate the suitability of the regularisation method to AVSR and encourage researchers to rethink the multimodal convergence problem when having one dominant modality.

A Point-Wise LiDAR and Image Multimodal Fusion Network (PMNet) for Aerial Point Cloud 3D Semantic Segmentation

3D semantic segmentation of point cloud aims at assigning semantic labels to each point by utilizing and respecting the 3D representation of the data. Detailed 3D semantic segmentation of urban areas can assist policymakers, insurance companies, governmental agencies for applications such as urban growth assessment, disaster management, and traffic supervision. The recent proliferation of remote sensing techniques has led to producing high resolution multimodal geospatial data. Nonetheless, currently, only limited technologies are available to fuse the multimodal dataset effectively. Therefore, this paper proposes a novel deep learning-based end-to-end Point-wise LiDAR and Image Multimodal Fusion Network (PMNet) for 3D segmentation of aerial point cloud by fusing aerial image features. PMNet respects basic characteristics of point cloud such as unordered, irregular format and permutation invariance. Notably, multi-view 3D scanned data can also be trained using PMNet since it considers aerial point cloud as a fully 3D representation. The proposed method was applied on two datasets (1) collected from the urban area of Osaka, Japan and (2) from the University of Houston campus, USA and its neighborhood. The quantitative and qualitative evaluation shows that PMNet outperforms other models which use non-fusion and multimodal fusion (observational-level fusion and feature-level fusion) strategies. In addition, the paper demonstrates the improved performance of the proposed model (PMNet) by over-sampling/augmenting the medium and minor classes in order to address the class-imbalance issues.

Multimodal and Multi-Model Deep Fusion for Fine Classification of Regional Complex Landscape Areas Using ZiYuan-3 Imagery

Land cover classification (LCC) of complex landscapes is attractive to the remote sensing community but poses great challenges. In complex open pit mining and agricultural development landscapes (CMALs), the landscape-specific characteristics limit the accuracy of LCC. The combination of traditional feature engineering and machine learning algorithms (MLAs) is not sufficient for LCC in CMALs. Deep belief network (DBN) methods achieved success in some remote sensing applications because of their excellent unsupervised learning ability in feature extraction. The usability of DBN has not been investigated in terms of LCC of complex landscapes and integrating multimodal inputs. A novel multimodal and multi-model deep fusion strategy based on DBN was developed and tested for fine LCC (FLCC) of CMALs in a 109.4 km2 area of Wuhan City, China. First, low-level and multimodal spectral–spatial and topographic features derived from ZiYuan-3 imagery were extracted and fused. The features were then input into a DBN for deep feature learning. The developed features were fed to random forest and support vector machine (SVM) algorithms for classification. Experiments were conducted that compared the deep features with the softmax function and low-level features with MLAs. Five groups of training, validation, and test sets were performed with some spatial auto-correlations. A spatially independent test set and generalized McNemar tests were also employed to assess the accuracy. The fused model of DBN-SVM achieved overall accuracies (OAs) of 94.74% ± 0.35% and 81.14% in FLCC and LCC, respectively, which significantly outperformed almost all other models. From this model, only three of the twenty land covers achieved OAs below 90%. In general, the developed model can contribute to FLCC and LCC in CMALs, and more deep learning algorithm-based models should be investigated in future for the application of FLCC and LCC in complex landscapes.

Indic handwritten script identification using offline-online multi-modal deep network

In this paper, we propose a novel approach of word-level Indic script identification using only character-level data in training stage. Our method uses a multi-modal deep network which takes both offline and online modality of the data as input in order to explore the information from both the modalities jointly for script identification task. We take handwritten data in either modality as input and the opposite modality is generated through intermodality conversion. Thereafter, we feed this offline-online modality pair to our network. Hence, along with the advantage of utilizing information from both the modalities, the proposed framework can work for both offline and online script identification which alleviates the need for designing two separate script identification modules for individual modality. We also propose a novel conditional multi-modal fusion scheme to combine the information from offline and online modality which takes into account the original modality of the data being fed to our network and thus it combines adaptively. An exhaustive experimental study has been done on a data set including English(Roman) and 6 other official Indic scripts. Our proposed scheme outperforms traditional classifiers along with handcrafted features and deep learning based methods. Experiment results show that using only character level training data can achieve competitive performance against traditional training using word level data.

A snapshot research and implementation of multimodal information fusion for data-driven emotion recognition

With the rapid development of artificial intelligence and mobile Internet, the new requirements for human-computer interaction have been put forward. The personalized emotional interaction service is a new trend in the human-computer interaction field. As a basis of emotional interaction, emotion recognition has also introduced many new advances with the development of artificial intelligence. The current research on emotion recognition mostly focuses on single-modal recognition such as expression recognition, speech recognition, limb recognition, and physiological signal recognition. However, the lack of the single-modal emotional information and vulnerability to various external factors lead to lower accuracy of emotion recognition. Therefore, multimodal information fusion for data-driven emotion recognition has been attracting the attention of researchers in the affective computing filed. This paper reviews the development background and hot spots of the data-driven multimodal emotion information fusion. Considering the real-time mental health monitoring system, the current development of multimodal emotion data sets, the multimodal features extraction, including the EEG, speech, expression, text features, and multimodal fusion strategies and recognition methods are discussed and summarized in detail. The main objective of this work is to present a clear explanation of the scientific problems and future research directions in the multimodal information fusion for data-driven emotion recognition field.

One-Shot SADI-EPE: A Visual Framework of Event Progress Estimation

In many practical engineering applications, the number of actions that have been finished should be known, particularly for an untrimmed video sequence that includes an event with a series of actions, it is important to know the number of actions that have been finished. In this paper, we termed this process as visual event progress estimation (EPE). However, the research related to this problem is few in the research community. To solve this problem, a visual human action analysis-based framework, namely one-shot simultaneously action detection and identification (SADI)-EPE, is presented in this paper. The visual EPE is modeled as an online one-shot learning-based problem; sliding window and attention-based bag of key poses formulate our framework. Unlike most of the action analysis methods relying on a number of training data of some predefined classes, our method can realize SADI for any event if one sample of the event is given, which makes it feasible for practical applications. At the same time, not only SADI but also the progress estimation of the event can be realized by our algorithm. In terms of methodology, the key pose is defined by an invariant pose descriptor from skeletal data and silhouette data. Moreover, in order to extract representative and discriminative poses from one training sample, we present a new bidirectional $k$ NN-based attention weighted key pose selection method, which can filter the unrelated actions and model different importance of various key poses. In addition, an attention-based multi-modal fusion scheme, which addresses the difficulty of high-dimensional features and few training samples, is proposed to augment the performance of our algorithm. Finally, we propose an evaluation criterion for the estimation problem. Extensive results demonstrated the efficacy of our proposed framework.

Fusing Deep Learning and Sparse Coding for SAR ATR

We propose a multimodal and multidiscipline data fusion strategy appropriate for automatic target recognition (ATR) on synthetic aperture radar imagery. Our architecture fuses a proposed clustered version of the AlexNet convolutional neural network with sparse coding theory that is extended to facilitate an adaptive elastic net optimization concept. Evaluation on the MSTAR dataset yields the highest ATR performance reported yet, which is 99.33% and 99.86% for the three- and ten-class problems, respectively.

Secure multimodal biometric system based on diffused graphs and optimal score fusion

Biometric authentication systems offer potential advantages over traditional knowledge-based methods. However, most of the biometric systems which are extensively used lack template security and robustness. In order to address these issues, in this study, the authors have proposed a multimodal biometric system based on the combination of multiple modalities and optimal score level fusion. In addition, key features are introduced for each modality for generating cancellable biometric features. Features from individual traits are combined with corresponding key features to provide feature transformation. A robust template is generated by diffusion of individual transformed matrices using graph-based random walk cross-diffusion. In addition, individual classifier score is optimally fused using proposed multistage score level fusion model. Optimal belief masses for individual classifier are determined using cuckoo search optimisation. Wherein, optimal classifier beliefs are fused using DSmT based proportional conflict redistribution (PCR-6) rules. Experimental results demonstrate that optimal score fusion applied on cross-diffused features produce better results than existing state-of-the-art multimodal fusion schemes. On average of the outcome, equal error rate and accuracy achieved using the proposed method on four chimeric benchmarked datasets, are 2.32 and 98.316%.

Multibiometric fusion strategy and its applications: A review

The unimodal biometric based system faced several inherent problems like lack of uniqueness, intra-class variation, non-universality, noisy data (presence of dirt on the sensor), restricted degree of freedom, unacceptable error rate, failure-to-enroll and spoofing attack. Multibiometric is one of the best choices to overcome these problems. Multibiometric fusion plays an important role to enhance the overall performance of the system, in which two or more individual biometric are combined together to form a better performance system. The proper use of fusion strategy is very important in the multibiometric system because it can affect the overall performance and accuracy level of the systems. In designing a multibiometric based system we can use various methods and fusion strategies to combine information from multiple sources. This paper is an in-depth study on multibiometric (multimodal, multialgorithm, multi-sample, multi-sensor and multi-instance) fusion strategy and its different applications. In addition, this paper also discusses the different methodology used in a fusion process (Sensor, Feature, Score, Decision, Rank) of multibiometric systems from last three decades and examines the methods used, to explore their successes and failure.

Integrating Both Visual and Audio Cues for Enhanced Video Caption

Video caption refers to generating a descriptive sentence for a specific short video clip automatically, which has achieved remarkable success recently. However, most of the existing methods focus more on visual information while ignoring the synchronized audio cues. We propose three multimodal deep fusion strategies to maximize the benefits of visual-audio resonance information. The first one explores the impact on cross-modalities feature fusion from low to high order. The second establishes the visual-audio short-term dependency by sharing weights of corresponding front-end networks. The third extends the temporal dependency to long-term through sharing multimodal memory across visual and audio modalities. Extensive experiments have validated the effectiveness of our three cross-modalities fusion strategies on two benchmark datasets, including Microsoft Research Video to Text (MSRVTT) and Microsoft Video Description (MSVD). It is worth mentioning that sharing weight can coordinate visual- audio feature fusion effectively and achieve the state-of-art performance on both BELU and METEOR metrics. Furthermore, we first propose a dynamic multimodal feature fusion framework to deal with the part modalities missing case. Experimental results demonstrate that even in the audio absence mode, we can still obtain comparable results with the aid of the additional audio modality inference module.

A quantum-inspired multimodal sentiment analysis framework

Multimodal sentiment analysis aims to capture diversified sentiment information implied in data that are of different modalities (e.g., an image that is associated with a textual description or a set of textual labels). The key challenge is rooted on the “semantic gap” between different low-level content features and high-level semantic information. Existing approaches generally utilize a combination of multimodal features in a somehow heuristic way. However, how to employ and combine multiple information from different sources effectively is still an important yet largely unsolved problem. To address the problem, in this paper, we propose a Quantum-inspired Multimodal Sentiment Analysis (QMSA) framework. The framework consists of a Quantum-inspired Multimodal Representation (QMR) model (which aims to fill the “semantic gap” and model the correlations between different modalities via density matrix), and a Multimodal decision Fusion strategy inspired by Quantum Interference (QIMF) in the double-slit experiment (in which the sentiment label is analogous to a photon, and the data modalities are analogous to slits). Extensive experiments are conducted on two large scale datasets, which are collected from the Getty Images and Flickr photo sharing platform. The experimental results show that our approach significantly outperforms a wide range of baselines and state-of-the-art methods.

Learning Multimodal Deep Representations for Crowd Anomaly Event Detection

Anomaly event detection in crowd scenes is extremely important; however, the majority of existing studies merely use hand-crafted features to detect anomalies. In this study, a novel unsupervised deep learning framework is proposed to detect anomaly events in crowded scenes. Specifically, low-level visual features, energy features, and motion map features are simultaneously extracted based on spatiotemporal energy measurements. Three convolutional restricted Boltzmann machines are trained to model the mid-level feature representation of normal patterns. Then a multimodal fusion scheme is utilized to learn the deep representation of crowd patterns. Based on the learned deep representation, a one-class support vector machine model is used to detect anomaly events. The proposed method is evaluated using two available public datasets and compared with state-of-the-art methods. The experimental results show its competitive performance for anomaly event detection in video surveillance.

Affect recognition from facial movements and body gestures by hierarchical deep spatio-temporal features and fusion strategy

Affect presentation is periodic and multi-modal, such as through facial movements, body gestures, and so on. Studies have shown that temporal selection and multi-modal combinations may benefit affect recognition. In this article, we therefore propose a spatio-temporal fusion model that extracts spatio-temporal hierarchical features based on select expressive components. In addition, a multi-modal hierarchical fusion strategy is presented. Our model learns the spatio-temporal hierarchical features from videos by a proposed deep network, which combines a convolutional neural networks (CNN), bilateral long short-term memory recurrent neural networks (BLSTM-RNN) with principal component analysis (PCA). Our approach handles each video as a “video sentence.” It first obtains a skeleton with the temporal selection process and then segments key words with a certain sliding window. Finally, it obtains the features with a deep network comprised of a video-skeleton and video-words. Our model combines the feature level and decision level fusion for fusing the multi-modal information. Experimental results showed that our model improved the multi-modal affect recognition accuracy rate from 95.13% in existing literature to 99.57% on a face and body (FABO) database, our results have been increased by 4.44%, and it obtained a macro average accuracy (MAA) up to 99.71%.

Skeleton-Based Action Recognition Using Spatio-Temporal LSTM Network with Trust Gates.

Skeleton-based human action recognition has attracted a lot of research attention during the past few years. Recent works attempted to utilize recurrent neural networks to model the temporal dependencies between the 3D positional configurations of human body joints for better analysis of human activities in the skeletal data. The proposed work extends this idea to spatial domain as well as temporal domain to better analyze the hidden sources of action-related information within the human skeleton sequences in both of these domains simultaneously. Based on the pictorial structure of Kinect's skeletal data, an effective tree-structure based traversal framework is also proposed. In order to deal with the noise in the skeletal data, a new gating mechanism within LSTM module is introduced, with which the network can learn the reliability of the sequential data and accordingly adjust the effect of the input data on the updating procedure of the long-term context representation stored in the unit's memory cell. Moreover, we introduce a novel multi-modal feature fusion strategy within the LSTM unit in this paper. The comprehensive experimental results on seven challenging benchmark datasets for human action recognition demonstrate the effectiveness of the proposed method.

Driver Missense Mutation Identification Using Feature Selection and Model Fusion.

Driver mutations propel oncogenesis and occur much less frequently than passenger mutations. The need for automatic and accurate identification of driver mutations has increased dramatically with the exponential growth of mutation data. Current computational solutions to identify driver mutations rely on sequence homology. Here we construct a machine learning-based framework that does not rely on sequence homology or domain knowledge to predict driver missense mutations. A windowing approach to represent the local environment of the sequence around the mutation point as a mutation sample is applied, followed by extraction of three sequence-level features from each sample. After selecting the most significant features, the support vector machine and multimodal fusion strategies are employed to give final predictions. The proposed framework achieves relatively high performance and outperforms current state-of-the-art algorithms. The ease of deploying the proposed framework and the relatively accurate performance make this solution applicable to large-scale mutation data analyses.

A multi-modal fusion scheme for the enhancement of PET/MR viewing.

PET/MR imaging offers the possibility to achieve in one-shot both functional information provided by PET imaging and morpho-functional information with excellent soft tissue contrast provided by MRI. As a result of a typical PET/MR acquisition, each voxel carries a large amount of multivariate information that can be combined into a single image for a synthetic multi-parametric viewing (image fusion). This work is aimed to explore and introduce a reliable fusion scheme able to blend the useful information carried out from each modality into a single meaningful image. The proposed approach consists of a transformation of the source images followed by a blending and a consecutive reconstruction of fused image into the original domain. Following this scheme, three different transformations of source images (wavelet domain, Fourier domain and identity) and two blending procedures (alpha-blending and gamma-blending) were considered. For a comprehensive assessment of the fusion schemes under investigation, the results were evaluated both qualitatively and quantitatively on a dataset of 60 naturally coregistered. FDG-PET/MR studies of different anatomical districts in presence of neurological as well as oncological findings. The quality of fused images was assessed by experts who visually evaluated the loss of useful information with respect to the original modalities. Quantitatively, the information loss was estimated by means of a boundaries-preservation based metric as well as a metric based on mutual information. Preliminary results show that the fusion scheme composed by a wavelet-domain transform and a gamma-blending better depicts useful information of the original modalities. In particular, the proposed approach allows to better unveil the MR signal underlying PET signal in fused images in oncological as well as neurological studies. Furthermore, the use of a gamma-blending procedure ensures the best performances in a fully automated manner.

Event oriented dictionary learning for complex event detection.

Complex event detection is a retrieval task with the goal of finding videos of a particular event in a large-scale unconstrained Internet video archive, given example videos and text descriptions. Nowadays, different multimodal fusion schemes of low-level and high-level features are extensively investigated and evaluated for the complex event detection task. However, how to effectively select the high-level semantic meaningful concepts from a large pool to assist complex event detection is rarely studied in the literature. In this paper, we propose a novel strategy to automatically select semantic meaningful concepts for the event detection task based on both the events-kit text descriptions and the concepts high-level feature descriptions. Moreover, we introduce a novel event oriented dictionary representation based on the selected semantic concepts. Toward this goal, we leverage training images (frames) of selected concepts from the semantic indexing dataset with a pool of 346 concepts, into a novel supervised multitask lp -norm dictionary learning framework. Extensive experimental results on TRECVID multimedia event detection dataset demonstrate the efficacy of our proposed method.

Complex Event Detection via Event Oriented Dictionary Learning

Complex event detection is a retrieval task with the goal of finding videos of a particular event in a large-scale unconstrained internet video archive, given example videos and text descriptions. Nowadays, different multimodal fusion schemes of low-level and high-level features are extensively investigated and evaluated for the complex event detection task. However, how to effectively select the high-level semantic meaningful concepts from a large pool to assist complex event detection is rarely studied in the literature. In this paper, we propose two novel strategies to automatically select semantic meaningful concepts for the event detection task based on both the events-kit text descriptions and the concepts high-level feature descriptions. Moreover, we introduce a novel event oriented dictionary representation based on the selected semantic concepts. Towards this goal, we leverage training samples of selected concepts from the Semantic Indexing (SIN) dataset with a pool of 346 concepts, into a novel supervised multi-task dictionary learning framework. Extensive experimental results on TRECVID Multimedia Event Detection (MED) dataset demonstrate the efficacy of our proposed method.