IPIARTI 2018

spsocmalaysia.org/ipiarti2018
_____________

COMMITTEE

The aim of stereo matching algorithm is to obtain the information of depth or distance to an input image. This is done by finding the matching pixel between two images at different viewpoints. The two-dimensional mapping of matching pixels is known as disparity map. This research topic has been studied for decades, which the depth from stereo remains a long- standing challenge. Several factors that make computations of stereo matching algorithm are challenging such as complex scenes, radiometric changes and repetition textures. The applications of depth stereopsis are intelligent vehicles, autonomous robotics and augmented reality. Stereo matching algorithms can be categorized into global and local methods. Global methods perform a matching process using global energy or a probability function over the whole image. The global methods involve high computational complexity and slow implementation. Therefore, it is not suitable for real-time applications. However, local methods solve the matching problem via a local analysis and aggregating matching costs over a support region at each pixel in the images. The local methods deliver fast execution and low computational requirement. The main challenge of stereo matching algorithm is to find the corresponding pixels in the low texture regions. These regions contain plain color pixels which make the corresponding process unable to determine the best matching pixels. Hence, this abstract proposes a new local-based stereo matching algorithm to improve the low texture regions which involves four stages. First, the matching cost function is developed using a combination of Absolute Differences (AD) and Gradient Matching (GM). Second, a new edge preserving filter is proposed at cost aggregation stage. This filter is known as hierarchical Guided Filter (GF) which is capable to increase the efficiency of preserving the object edges. Then, the optimization step uses a Winner-Take-All (WTA) strategy. The WTA strategy absorbs the minimal aggregated corresponding value for each valid pixel. Finally, the post-processing stage which is to refine the final disparity map. The unwanted and invalid pixels are still occurring at the occlusion and untextured areas. These unwanted pixels will be detected by Left-Right (LR) consistency checking process. Then, the fill-in process is carried out to replace the invalid pixels with a valid minimum pixel value. The disparity refinement step consists of implementing the weighted bilateral filter to remove the remaining noise which usually occurs during the fill-in process. The undirected graph segmentation and least square plane fitting process are used at the final step to recover the low texture regions on the final disparity map. Based on the experimental results using a standard benchmarking from the Middlebury dataset, the proposed algorithm is able to reduce the error and increase the accuracy against the low textured areas with 35.25% of noise reduction based on average of all error attribute. The proposed algorithm also produces good results on the real stereo images from the KITTI dataset.

Food Intake Gesture Monitoring System Based-On Depth Sensor

Food intake gesture technology is one of a new strategy for obesity people managing their health care while saving their time and money. Moreover, Obesity continues to be a serious public health concern in Malaysia and continuing to rise. Most of the dietary approach is not tracking and detecting the right calorie intake for weight loss, but currently used tools such as food diaries require users to manually record and track the food calories, making them difficult for daily use. We will be developing a new tool that counts the food intake bite by monitoring eating motion movement of caloric intake. This approach involves combining face and hand joint point for monitoring food intake of a user using Kinect Xbox One camera sensor. Rather than counting calories, scientists at Brigham Young University found dieters who eager to reduce their number of daily bites by 20 to 30 percent lost around two kilograms a month, regardless of what they eat. Research studies showed that most of the methods used to count bite are worn type devices which has high false alarm ratio. Today trend is going toward the non-wearable device. This sensor is used to capture skeletal data and face point cloud of user while eating and train the data to capture the motion and movement while eating. There is specific joint to be capture such as Jaw face point and wrist roll joint. Pitch, roll and yaw of hand wrist will be analyse to capture the hand rotation using joint quaternion of hand wrist bone. This system can help people who are trying to follow a proper way to reduce overweight or eating disorders by monitoring their meal intake and controlling eating rate.

A Fully Adaptive Image Classification Approach for Industrial Revolution 4.0

Industrial Revolution (IR) improves the way we live, work and interact with each other by using state of the art technologies. IR-4.0 describes a future state of industry which is characterized through the digitization of economic and production flows. The nine pillars of IR-4.0 are dependent on Big Data Analytics, Artificial Intelligence, Cloud Computing Technologies and Internet of Things (IoT). Image datasets are most valuable among other types of Big Data. Image Classification Models (ICM) are considered as an appropriate solution for Business Intelligence. However, due to complex image characteristics, one of the most critical issues encountered by the ICM is the Concept Drift (CD). Due to CD, ICM are not able to adapt and result in performance degradation in terms of accuracy. Therefore, ICM need better adaptability to avoid performance degradation during CD. Adaptive Convolutional ELM (ACNNELM) is one of the best existing ICM for handling multiple types of CD. However, ACNNELM does not have sufficient adaptability. This paper proposes a more autonomous adaptability module, based on Meta-Cognitive principles, for ACNNELM to further improve its performance accuracy during CD. The Meta-Cognitive module will dynamically select different CD handling strategies, activation functions, number of neurons and restructure ACNNELM as per changes in the data.
This research contribution will be helpful for improvement in various practical applications areas of Business Intelligence which are relevant to IR-4.0 and TN50 (e.g., Automation Industry, Autonomous Vehicle, Expert Agriculture Systems, Intelligent Education System, and Healthcare etc.)

Augmenting the Perception of the Blind and Low Vision to Increase their Levels of Independent Navigation through Wearable Technology

This report presents the research of an assistive technology to help the blind and low vision people (BLVs) in navigating their pathway. The research addresses specifically the challenge of negotiating surface discontinuities typically found within urban areas. Based on a design science research process, a lightweight, small and unobtrusive wearable prototype is developed for the study. Tailored for a computer vision task, the prototype is equipped with a tiny stereo camera and a mobile computer. With this prototype, a set of video data that exemplifies the issue of surface discontinuity was collected from the field. These data were used to train a machine learning model to classify the condition of the pathway. A purpose-built tri-channel-single-input stacked convolutional neural network was assembled, experimented, trained and optimized across a series of configuration and hyperparameter setting. With the incorporation of the best trained model, the similar prototype is then repurposed such that it is working in real-time to classify the condition of a pathway with some simple feedback to the user. The efficacy of the prototype to classify the surface condition was then evaluated. The optimized model was able to achieve classification accuracy of 96.7% at field tests.

Image Processing, Image Analysis and Real-Time Imaging (IPIARTI) Symposium 2018
Program Keynotes & Invited Talks Venue Committee Organizer
ABSTRACT SUBMISSION FOR TECHNICAL PRESENTATIONS Prospective presenters are invited to submit a one-page abstract of their work. The selected presenters will received a certificate of appreciation as invited speaker. DEADLINE: 30 June 2018 Email abstracts to koksheik.wong.my@ieee.org ------------------------ Official REGISTRATION (with refreshments) is open. Click here to register.
12 July 2018 (Thursday) @ Auditorium 1, Building 5 Monash University Malaysia Jalan Lagoon Selatan, 47500 Bandar Sunway Selangor Darul Ehsan, Malaysia
IEEE Signal Processing Society Malaysia Chapter is jointly organizing the 9^thSymposium on Image Processing, Image Analysis and Real-Time Imaging (IPIARTI 2018) with the IEEE Monash Student Branch. This FREE annual event, open to all members and non-members of the IEEE, is organized with the objectives of: · to bring the university and industry community together to share and discuss the latest trends in image processing, analysis and real-time implementation, and · to promote the IEEE Signal Processing Society Malaysia Chapter to the academic and industry community in Malaysia as a forum for professional networking and advancement. Program 09.00 - 09.30: Registration 09.30 - 09.45: Welcome Address Assoc. Prof. Dr. Wong Kok Sheik Chair, IPIARTI 2018 & ExComm, IEEE Signal Processing Society Malaysia Chapter Prof. Anthony Guo Head, School of Enginering Head, School of Information Technology. Monash University Malaysia, Malaysia 09.45 - 10.30: Keynote Speech #1 "Applications of Image and Video analysis for deblurring and heart rate estimation" Prof. Dr. Raveendran Paramesran Faculty of Engineering, University of Malaya. 10.30 - 10.50: Coffee Break 10.50 - 11.40: Keynote Speech #2 "Computer Vision Applications for IoT Solutions" Dr . Minemura Kazuki Internet of things group / architecture silicon and platform engineering, Intel Malaysia 11.40 - 12.30: Keynote Speech #3 "Deep Image Understanding" Assoc. Prof. Dr .Chan Chee Seng Head, Department of Artificial Intelligence, Faculty of Computer Science and Information Technology, University of Malaya. 12:30 - 12:45: IEEE SPS Malaysia Research Excellence Awards 12.45 - 13.00: Membership & Senior Member Elevation Drive / Announcements 13.00 - 14.00: Lunch Break 14.00 - 16.00: Technical Presentations / Invited Talks (in room 6-3-15) "Local-based StereoMatching Algorithm for Improvement of Low Texture Region" Dr. Rostam Affendi Hamzah (UTEM) "Food Intake Gesture Monitoring System Based-On Depth Sensor" Muhammad fuad bin kassim (UTHM) "A Fully Adaptive Image Classification Approach for Industrial Revolution 4.0" Syed Muslim Jameel (UTP) "Augmenting the Perception of the Blind and Low Vision to Increase their Levels of Independent Navigation through Wearable Technology" Leong Kuan Yew (Monash) 16.00 - 16.30: Certificate Presentation & Closing + Refreshment		spsocmalaysia.org/ipiarti2018 _____________ COMMITTEE Chair: Wong Kok Sheik IEEE SPS Malaysia Chapter / Monash University Malaysia Local Arrangement : IEEE Monash University- Sunway Campus Student Branch Committee Members: Syed Abdul Rahman Syed Abu Bakar Universiti Teknologi Malaysia (UTM) Mohammad Faizal Ahmad Fauzi Multimedia University (MMU) Nor’aini Abdul Jalil Wavesmiles Syed Khaleel Ahmed Universiti Tenaga Nasional (UNITEN) Vijanth Sagayan Asirvadam Universiti Teknologi PETRONAS (UTP) Sabira Khatun Universiti Malaysia Pahang (UMP) Hezerul Abdul Karim Multimedia University (MMU) Kushsairy Abdul Kadir Universiti Kuala Lumpur (UniKL) Rajasvaran Logeswaran Asia Pacific University of Technology & Innovation (APU) Mohd Norzali Haji Mohd Universiti Tun Hussein Onn Malaysia (UTHM) ABSTRACT Submission FOR TECHNICAL PRESENTATIONS Prospective presenters are invited to submit a one-page abstract of their work. The selected presenters will receive a certificate of appreciation as an invited speaker. DEADLINE: email abstracts to koksheik.wong.my@ieee.org by 21 Jun 2018 REGISTRATION Whether presenting or attending the symposium, please sign-up (for logistics) by clicking here by 30 June 2018 10 July 2018
KEYNOTES
KEYNOTE 1 Prof. Dr. Raveendhran Paramesran Faculty of Engineering, University of Malaya.	Applications of Image and Video Analysis for Deblurring and Heart Rate Estimation With the increasing popularity of digital cameras and the internet in the 21st century, digital images play an important role in human life. However, digital photos taken from cameras maybe have various distortion types. One of them is blur, which includes the motion blur (due to camera movement or camera shake, object movement and slow shutter speed), and the out-of-focused blur (incorrect camera auto-focusing or focusing at the wrong subject of interest). Efficient image quality assessment algorithms, therefore, are essential for the selection of the digital images for final production in photo-books, posters and magazines to ensure that they are above the minimum quality level. Image quality assessment (IQA), at least to some extent, is still based on human observers׳ opinions because we are the ultimate end-users. Based on the characteristics of the human visual system (HVS), various methods have been proposed to evaluate the image quality objectively to be in line with the subjective human evaluation. One issue that has challenged researchers in image processing field is the usage of pattern features for image quality assessment. An important characteristic in choosing features for No-Reference (NR) IQA is that they should closely exhibit human visual system (HVS) perception of image blur distortion. One particular moment that interests us is Exact Zernike moments (EZMs). In this presentation, we first show a set of EZMs that closely exhibit human quality scores for images of various degrees of blurriness. Inspired by these advantages, a no-reference blur metric is formulated using the proposed EZMs and gradient magnitude (GM) which acts as a weight to encode the contrast information and a support vector regressor (SVR) is trained to assign a quality score. In the second part of the presentation, we will show how heart-rate can be estimated from facial images in the video sequences. The contraction and relaxation of heart muscles during each cardiac cycle alters the colour variations of the face. The heart rate can be obtained analysing this color change of the face using robust image analysis.
KEYNOTE 2 Dr. Minemura Kazuki Internet of things group/architecture silicon and platform engineering, Intel Malaysia	Computer Vision Applications for IoT Solution The advancement in network and computing technology has enabled seamless integration of smart devices, sensors, embedded electronics. The amount of data exchanges in the network has grown significantly, enabling opportunities to analyze the data for smarter decision making. New application and usage model has been created. It is called internet of things (IOT). In IoT solution, vision data and its applications have significant meaning, because cameras and video are becoming not only lower cost device but also getting more ubiquitous. As a matter of fact, the number of video applications is growing across industries and the evolution of video is enabling real-time environment perception and increasing the amount of data. Computer vision applications are software solution and have been being researched and developed in all major IoT markets, such as, transportation, public services, retail, home and buildings, factories, healthcare and even drones. In this talk, we first show a framework of face recognition end-end solution. This framework consists of three applications, e.g. smart camera (including face detection, face landmark detection, and face alignment), feature extraction, face matching, and it enables to handle massive face database and to process parallel front matching by multi-core programming. In the second part of the talk, we will show how real-time multiclass object detection with 3D Lidar point cloud can be achieved from computer vision perspective. We achieved execution times as fast as 50 FPS using desktop GPUs, and up to 10 FPS on a single Intel Core i5 CPU.
KEYNOTE 3 Assoc. Prof. Dr. Chan Chee Seng Head, Department of Artificial Intelligence, Faculty of Computer Science and Information Technology, University of Malaya.	Deep Image Understanding This talk will present a study from large-scale classification of fine-art paintings using the Deep Convolutional Network, to synthetically generate of them using an extension of the Generative Adversarial Networks (GANs), namely as ARTGAN. The objectives of this research are two-folds. On one hand, we would like to train an end-to-end deep convolution model to investigate the capability of the deep model in fine-art painting classification problem. We argue that classification of fine-art collections is a more challenging problem in comparison to objects or face recognition. This is because some of the artworks are non-representational nor figurative, and might requires imagination to recognize them. Hence, a question arose is that does a machine have or able to capture “imagination” in paintings? One way to find out is train a deep model and then visualize the low-level to high-level features learnt. In the experiment, we employed the recently publicly available large-scale “Wikiart paintings” dataset that consists of more than 80,000 paintings and our solution achieved state-of-the-art results in overall performance. On the other hand, we would like realize if machine can synthetically generate challenging and complex images such as artwork that have abstract characteristics. Empirically, it shows that ARTGAN is capable to create realistic artwork, as well as generate compelling real world images that globally look natural with clear shape on CIFAR-10. The source code and models are available at https://github.com/cs-chan/Artwork-Synthesis-Classification
TECHNICAL PRESENTATIONS
INVITED TALK #1 Dr. Rostam Affendi Hamzah M. Saad Hamid, A. F. Kadmin, and S. Fakhar Abd Gani Fakulti Teknologi Kejuruteraan, Universiti Teknikal Malaysia Melaka	Local-based StereoMatching Algorithm for Improvement of Low Texture Region The aim of stereo matching algorithm is to obtain the information of depth or distance to an input image. This is done by finding the matching pixel between two images at different viewpoints. The two-dimensional mapping of matching pixels is known as disparity map. This research topic has been studied for decades, which the depth from stereo remains a long- standing challenge. Several factors that make computations of stereo matching algorithm are challenging such as complex scenes, radiometric changes and repetition textures. The applications of depth stereopsis are intelligent vehicles, autonomous robotics and augmented reality. Stereo matching algorithms can be categorized into global and local methods. Global methods perform a matching process using global energy or a probability function over the whole image. The global methods involve high computational complexity and slow implementation. Therefore, it is not suitable for real-time applications. However, local methods solve the matching problem via a local analysis and aggregating matching costs over a support region at each pixel in the images. The local methods deliver fast execution and low computational requirement. The main challenge of stereo matching algorithm is to find the corresponding pixels in the low texture regions. These regions contain plain color pixels which make the corresponding process unable to determine the best matching pixels. Hence, this abstract proposes a new local-based stereo matching algorithm to improve the low texture regions which involves four stages. First, the matching cost function is developed using a combination of Absolute Differences (AD) and Gradient Matching (GM). Second, a new edge preserving filter is proposed at cost aggregation stage. This filter is known as hierarchical Guided Filter (GF) which is capable to increase the efficiency of preserving the object edges. Then, the optimization step uses a Winner-Take-All (WTA) strategy. The WTA strategy absorbs the minimal aggregated corresponding value for each valid pixel. Finally, the post-processing stage which is to refine the final disparity map. The unwanted and invalid pixels are still occurring at the occlusion and untextured areas. These unwanted pixels will be detected by Left-Right (LR) consistency checking process. Then, the fill-in process is carried out to replace the invalid pixels with a valid minimum pixel value. The disparity refinement step consists of implementing the weighted bilateral filter to remove the remaining noise which usually occurs during the fill-in process. The undirected graph segmentation and least square plane fitting process are used at the final step to recover the low texture regions on the final disparity map. Based on the experimental results using a standard benchmarking from the Middlebury dataset, the proposed algorithm is able to reduce the error and increase the accuracy against the low textured areas with 35.25% of noise reduction based on average of all error attribute. The proposed algorithm also produces good results on the real stereo images from the KITTI dataset.
INVITED TALK #2 Muhammad Fuad bin Kassim and Dr. Mohd Norzali Bin Hj Mohd Universiti Tun Hussein Onn Malaysia	Food Intake Gesture Monitoring System Based-On Depth Sensor Food intake gesture technology is one of a new strategy for obesity people managing their health care while saving their time and money. Moreover, Obesity continues to be a serious public health concern in Malaysia and continuing to rise. Most of the dietary approach is not tracking and detecting the right calorie intake for weight loss, but currently used tools such as food diaries require users to manually record and track the food calories, making them difficult for daily use. We will be developing a new tool that counts the food intake bite by monitoring eating motion movement of caloric intake. This approach involves combining face and hand joint point for monitoring food intake of a user using Kinect Xbox One camera sensor. Rather than counting calories, scientists at Brigham Young University found dieters who eager to reduce their number of daily bites by 20 to 30 percent lost around two kilograms a month, regardless of what they eat. Research studies showed that most of the methods used to count bite are worn type devices which has high false alarm ratio. Today trend is going toward the non-wearable device. This sensor is used to capture skeletal data and face point cloud of user while eating and train the data to capture the motion and movement while eating. There is specific joint to be capture such as Jaw face point and wrist roll joint. Pitch, roll and yaw of hand wrist will be analyse to capture the hand rotation using joint quaternion of hand wrist bone. This system can help people who are trying to follow a proper way to reduce overweight or eating disorders by monitoring their meal intake and controlling eating rate.
INVITED TALK #3 Syed Muslim Jameel Manzoor Ahmed Hashmani, Hitham Alhussain and Arif Budiman Universiti Teknologi PETRONAS, Sri Iskandar, Perak, Malaysia High Performance Cloud Computing Center (HPC3)	A Fully Adaptive Image Classification Approach for Industrial Revolution 4.0 Industrial Revolution (IR) improves the way we live, work and interact with each other by using state of the art technologies. IR-4.0 describes a future state of industry which is characterized through the digitization of economic and production flows. The nine pillars of IR-4.0 are dependent on Big Data Analytics, Artificial Intelligence, Cloud Computing Technologies and Internet of Things (IoT). Image datasets are most valuable among other types of Big Data. Image Classification Models (ICM) are considered as an appropriate solution for Business Intelligence. However, due to complex image characteristics, one of the most critical issues encountered by the ICM is the Concept Drift (CD). Due to CD, ICM are not able to adapt and result in performance degradation in terms of accuracy. Therefore, ICM need better adaptability to avoid performance degradation during CD. Adaptive Convolutional ELM (ACNNELM) is one of the best existing ICM for handling multiple types of CD. However, ACNNELM does not have sufficient adaptability. This paper proposes a more autonomous adaptability module, based on Meta-Cognitive principles, for ACNNELM to further improve its performance accuracy during CD. The Meta-Cognitive module will dynamically select different CD handling strategies, activation functions, number of neurons and restructure ACNNELM as per changes in the data. This research contribution will be helpful for improvement in various practical applications areas of Business Intelligence which are relevant to IR-4.0 and TN50 (e.g., Automation Industry, Autonomous Vehicle, Expert Agriculture Systems, Intelligent Education System, and Healthcare etc.)
INVITED TALK #4 Leong Kuan Yew School of Information Tecehnology Monash University Malaysia	Augmenting the Perception of the Blind and Low Vision to Increase their Levels of Independent Navigation through Wearable Technology This report presents the research of an assistive technology to help the blind and low vision people (BLVs) in navigating their pathway. The research addresses specifically the challenge of negotiating surface discontinuities typically found within urban areas. Based on a design science research process, a lightweight, small and unobtrusive wearable prototype is developed for the study. Tailored for a computer vision task, the prototype is equipped with a tiny stereo camera and a mobile computer. With this prototype, a set of video data that exemplifies the issue of surface discontinuity was collected from the field. These data were used to train a machine learning model to classify the condition of the pathway. A purpose-built tri-channel-single-input stacked convolutional neural network was assembled, experimented, trained and optimized across a series of configuration and hyperparameter setting. With the incorporation of the best trained model, the similar prototype is then repurposed such that it is working in real-time to classify the condition of a pathway with some simple feedback to the user. The efficacy of the prototype to classify the surface condition was then evaluated. The optimized model was able to achieve classification accuracy of 96.7% at field tests.
VENUE / LOCATION
Monash University Malaysia is located in Bandar Sunway (see map on the right). It is accessible by public transportation. The nearest BRT station is called "SunU-Monash BRT station". Click here for Googe map.		Auditorium 1, Building 5 is on the ground level. Just walk straight towards the LIBRARY from the MAIN GATE (see image at the top of this page). Auditorium 1 is just next to the coffee stall (The Mad Alchemy @ Monash).
Parking: There are 2 public parking areas: (A) Student car park [click here] RM2 per entry (B) BRT Sunway Depot Hour rate (max RM6 per day)		Enquiries: For assistance, including finding your way to the location, please email koksheik.wong.my@ieee.org.
Back to Top
SPONSORED BY:

CO-ORGANIZED BY:
IEEE Monash University- Sunway Campus Student Branch
Back to Top
© 2018 IEEE Signal Processing Society Malaysia Chapter. All rights reserved.