There is a common tendency in object recognition research to accumulate large volumes of image features to improve performance. However, whether using more information contributes to higher accuracy is still controversial given the increased computational cost. This work investigates the performance of grayscale images compared to RGB counterparts for visual object classification. A comparison between object recognition based on RGB images and RGB images converted to grayscale was conducted using a cascaded CNN-RNN neural network structure, and compared with other types of commonly used classifiers such as Random Forest, SVM and SP-HMP. Experimental results showed that classification with grayscale images resulted in higher accuracy classification than with RGB images across the different types of classifiers. Results also demonstrated that utilizing a small receptive field CNN and edgy feature selection on grayscale images can result in higher classification accuracy with the advantage of reduced computational cost.

Deep neural networks (DNNs) trained on large data sets have been shown to be able to capture high-quality features describing image data. Numerous studies have proposed various ways to transfer DNN structures trained on large data sets to perform classification tasks represented by relatively small data sets. Due to the limitations of these proposals, it is not well known how to effectively adapt the pre-trained model into the new task. Typically, the transfer process uses a combination of fine-tuning and training of adaptation layers; however, both tasks are susceptible to problems with data shortage and high computational complexity. This paper proposes an improvement to the well-known AlexNet feature extraction technique. The proposed approach applies a recursive neural network structure on features extracted by a deep convolutional neural network pre-trained on a large data set. Object recognition experiments conducted on the Washington RGBD image data set have shown that the proposed method has the advantages of structural simplicity combined with the ability to provide higher recognition accuracy at a low computational cost compared with other relevant methods. The new approach requires no training at the feature extraction phase, and can be performed very efficiently as the output features are compact and highly discriminative, and can be used with a simple classifier in object recognition settings.

There has been much recent interest, both from industry and research communities, in 3D video technologies and processing techniques. However, with the standardisation of 3D video coding well underway and researchers studying 3D multimedia delivery and users' quality of multimedia experience in 3D video environments, there exist few publicly available databases of 3D video content. Further, there are even fewer sources of uncompressed 3D video content for flexible use in a number of research studies and applications. This paper thus presents a preliminary version of RMIT3DV: an uncompressed HD 3D video database currently composed of 31 video sequences that encompass a range of environments, lighting conditions, textures, motion, etc. The database was natively filmed on a professional HD 3D camera, and this paper describes the 3D film production workflow in addition to the database distribution and potential future applications of the content. The database is freely available online via the creative commons license, and researchers are encouraged to contribute 3D content to grow the resource for the (HD) 3D video research community.

Multimedia is ubiquitously available online with large amounts of video increasingly consumed through Web sites such as YouTube or Google Video. However, online multimedia typically limits users to visual/auditory stimulus, with onscreen visual media accompanied by audio. The recent introduction of MPEG-V proposed multi-sensory user experiences in multimedia environments, such as enriching video content with so-called sensory effects like wind, vibration, light, etc. In MPEG-V, these sensory effects are represented as Sensory Effect Metadata (SEM), which is additionally associated to the multimedia content. This paper presents three user studies that utilize the sensory effects framework of MPEG-V, investigating the emotional response of users and enhancement of Quality of Experience (QoE) of Web video sequences from a range of genres with and without sensory effects. In particular, the user studies were conducted in Austria and Australia to investigate whether geography and cultural differences affect users' elicited emotional responses and QoE.

The accurate tracking of zebrafish larvae movement is fundamental to research in many biomedical, pharmaceutical, and behavioral science applications. However, the locomotive characteristics of zebrafish larvae are significantly different from adult zebrafish, where existing adult zebrafish tracking systems cannot reliably track zebrafish larvae. Further, the far smaller size differentiation between larvae and the container render the detection of water impurities inevitable, which further affects the tracking of zebrafish larvae or require very strict video imaging conditions that typically result in unreliable tracking results for realistic experimental conditions. This paper investigates the adaptation of advanced computer vision segmentation techniques and multiple object tracking algorithms to develop an accurate, efficient and reliable multiple zebrafish larvae tracking system. The proposed system has been tested on a set of single and multiple adult and larvae zebrafish videos in a wide variety of (complex) video conditions, including shadowing, labels, water bubbles and background artifacts. Compared with existing state-of-the-art and commercial multiple organism tracking systems, the proposed system improves the tracking accuracy by up to 31.57% in unconstrained video imaging conditions. To facilitate the evaluation on zebrafish segmentation and tracking research, a dataset with annotated ground truth is also presented. The software is also publicly accessible.