Guest Editorial: Learning Multimedia for Real World Applications
- 834 Downloads
Multimedia data, as vivid and comprehensive, exist everywhere in our daily lives, in communication, education, manufacturing and service industries, and so on. Thus, how to improve the learning of multimedia data for real world applications has attracted widespread interests in the academy circle. This issue consists of 12 papers, which are briefly discussed as follows.
Three out of these 12 papers focus on developing feature representation to improve the performance of image retrieval, near duplicate image detection and video classification respectively. Sketch-based Image Retrieval (SBIR), which uses simple edge or contour images, is one important branch of Content-based Image Retrieval. However, SBIR is more difficult than CBIR due to the lack of visual information, this makes the Bag-of-Words (BoW) or codebook in SBIR hard to construct. The paper entitled “Sketch4Image: A Novel Framework for Sketch-Based Image Retrieval Based on Product Quantization with Coding Residuals” ( 10.1007/s11042-015-2645-y) proposes a novel SBIR framework based on Product Quantization (PQ) with sparse coding (SC) to construct an optimized codebook. In “Efficient Near-Duplicate Image Detection with a Local-Based Binary Representation” ( 10.1007/s11042-015-2472-1), Local-based Binary Representation is presented to encode an image as a binary vector for online near-duplicate image detection. The proposed representation is efficient to compute as well as robust in performance. It is also highly compact and it does not require any training phase. “A Bag-of-regions Representation for Video Classification” ( 10.1007/s11042-015-2876-y) proposes a bag-of-regions representation for high-level sparse representation of a video sequence. The BoR representation of a video sequence is obtained by extracting regions that exist in the majority of its frames and largely correspond to a single object.
Two papers study on the application in transportation, one is visual railway detection, and the other is traffic anomaly detection. “Visual Railway Detection by Superpixel based Intracellular Decisions” ( 10.1007/s11042-015-2654-x) proposes to detect railways based on superpixels. An SVM classifier is learned based on features, on which a TF-IDF like transform is applied, and it greatly improves the performance of the classification. “Traffic Anomaly Detection Based on Image Descriptor in Videos” ( 10.1007/s11042-015-2637-y) introduces a new traffic anomaly detection algorithm based on image description technology. The experimental results show that the proposed algorithm has improved performances of anomaly detection on both intersection traffic videos and main road traffic videos.
Two papers are related to social media, one is for sentiment analysis of social network multimedia, and the other is photo collection summarization. To address sentiment analysis of micro-blogging content, such as Twitter short messages, “A Multimodal Feature Learning Approach for Sentiment Analysis of Social Network Multimedia” ( 10.1007/s11042-015-2646-x) investigates the use of a multimodal feature learning approach, using neural network based models such as Skip-gram and Denoising Autoencoders. For photo collection summarization, the existing methods mainly consider the low-level features for photo representation only, while ignore many other useful features. The paper entitled “Multi-modal and Multi-scale Photo Collection Summarization” ( 10.1007/s11042-015-2658-6) proposes a multi-modal and multi-scale photo collection summarization method by leveraging multi-modal features, including time, location and high-level semantic features. The key photo ranking algorithm also takes the importance of both events and photos into consideration, and the proposed method allows users to control the scale of event segmentation and number of key photos selected.
One paper raises an interesting discussion on the importance of location information in saliency detection. “How Important is Location Information in Saliency Detection of Natural Images” ( 10.1007/s11042-015-2875-z) provides direct and quantitative analysis of the importance of location information for saliency detection in natural images. A location based saliency detection approach is proposed to completely initialize saliency maps with location information and propagate saliency among patches based on color similarity. The proposed method can handle natural images with different object positions and multiple salient objects.
“Compressive Sensing Reconstruction for Compressible Signal Based on Projection Replacement” ( 10.1007/s11042-015-2578-5) proposes the projection replacement (PR) algorithm by building the measurement space and its orthogonal complement space with singular value decomposition, and replacing the projection in measurement space of the reconstructed result with the pseudo-inverse one. The proposed PR algorithm eliminates the hypothetic measurement error in OMP and TSW-CS reconstructed model, and it guarantees theoretically that the PR results have a smaller error. Its effectiveness is verified experimentally with OMP and TSW-CS. The proposed algorithm can serve as a good reconstruction algorithm for the CS-based applications such as image coding, super-resolution, video retrieval etc.
In image denoising, we accepted the paper entitled “Non-local Sparse Regularization Model with Application to Image Denoising” ( 10.1007/s11042-015-2471-2), which studies on the denoising of natural images corrupted by Gaussian white noise. A framework is proposed to explore two sets of ideas involving on the one hand, locally learning a dictionary and estimating the sparse regularization signal descriptions for each coefficient; and on the other hand, nonlocally enforcing the invariance constraint by introducing patch self-similarities of natural images into the cost functional. The proposed framework outperforms the state-of-the-art, and makes it possible to effectively restore raw images from digital cameras at a reasonable speed and memory cost.
“An Implicit Relevance Feedback Method for CBIR with Real-Time Eye Tracking” ( 10.1007/s11042-015-2873-1) proposes a novel image retrieval system with implicit relevance feedback, named eye tracking based relevance feedback system (ETRFs). ETRFs is composed of three main modules: image retrieval subsystem based on bag-of-word architecture; user relevance assessment that implicitly acquires relevant images with the help of a modern eye tracker; and relevance feedback module that applies a weighted query expansion method to fuse users’ relevance feedback. ETRFs is implemented online and real time, which makes it remarkably distinguish from other online systems.
Last but not the least, we also accepted one paper studying on music. “Note Onset Detection Based on Sparse Decomposition” ( 10.1007/s11042-015-2656-8) studies on music onset detection, which is significant and essential for obtaining the high-level music features such as rhythm, beat, music paragraph and structure. This paper proposes a new algorithm for note onset detection based on sparse decomposition to solve the problem of lacking adaptiveness for representing the stationary and non-stationary part of the music signal in traditional onset detection methods. This is the first attempt to employ sparse decomposition into music onset detection.