Visual Inspection and Computational Vision

Contact: Professor Heikki Kälviäinen

Visual inspection and computational vision research at MVPR is focused on imaging, image processing, and image analysis methods for machine vision applications. The research topics include visual quality assessment, image-based process control, human quality perception and machine vision measurements, and visual object categorization. The applications are focused on industrial machine vision, especially the sawmill and papermaking industry, image-content based solutions, medical image processing, and photo-ID and animal biometrics for conservation of nature. One of the main objectives in forest industry applications is resource-efficient and environmentally sound production with the known quality, using less raw material, water, and energy.

Featured projects


Leap of Digitalisation for Sawmill Industry

The goal of this project is to develop the sawmill industry significantly via digitalisation. One of the main objectives is to build an information system connecting all the production steps of the mill so that it is possible to track the life cycle of raw material from the beginning to the end product. Raw material and its characteristics become individually traceable (e.g., logs and their parts) in the real-time industrial environment. This makes the use of raw material more efficient. At the same time, resource-efficient processes become more sustainable which benefits conservation of nature, especially in controlling the climate change. Modern sensor systems, new computational methods, and more robust mathematical models enable developing novel machine vision based measurements for quality control systems. The capabilities of the systems can be extended by using measured information from the sawmill processes and raw material. The obtained quality information can be connected via feedback loops to sorting the logs and optimizing the sawing processes. The developed innovations will be tested in the real industrial environment. The project includes international collaboration with the Computer Graphics Research Group at Brno University of Technology and Uncertainty Quantification Group at Massachusetts Institute of Technology (MIT). Contact Prof. Heikki Kälviäinen for futher information.


Automatic segmentation of overlapping objects for cell image analysis

In industrial and biomedical computer vision applications it is very important to deal with digital images which contain a large amount of overlapping objects. This leads to very useful applications such as cell imaging in the healthcare for diagnosis of cancer and nanoparticle imaging in the industry for quality control and material research. Our goal is to study automatic segmentation of overlapping objects. Manual segmentation is laborious, slow, and sensitive to interpretations. Images may contain tens of thousands of objects and it is time-consuming to detect attached and overlapping objects accurately. Recent developments in machine learning, especially with deep neural networks have enabled massive computations with large datasets. This project continues the work that was done in the COMPHI project where the state-of-the-art methods for nanoparticle segmentation were developed for industrial purposes. The main aim of this project is to extend our approach to cell imaging for the modern digital healthcare. The project includes international collaboration with the Laboratory for Image and Video Engineering (LIVE) at The University of Texas at Austin (UT) led by Professor Alan C. Bovik. Contact Prof. Heikki Kälviäinen for futher information.


Detection of overlapping objects

The goal is to analyze overlapping objects in an image, usually at microscopic level and for industrial and medical purposes. The task consists of the following research topics: to detect objects which are possibly attached and/or overlapping, to segment and to classify the objects, and to present the size distributions of the objects (absolute/relative). This project is a part of LUT Center of Excellence in Research, called Computational Photonics Imaging (COMPHI), in co-operation with LUT Applied Mathematics Laboratory, being associated with Academy of Finland's Center of Excellence in Research (CoE) in Inverse Modeling and Imaging. Currently, the COMPHI research is done in the CellVision project where University of Texas at Austin (UTA) is the international partner. Contact Prof. Heikki Kälviäinen for futher information.


Copyright: University of Eastern Finland (UEF)

Automatic image-based identification of Saimaa ringed seals

The Saimaa ringed seal (Phoca hispida saimensis) is one of the most endangered seals in the world. It is a symbol of Lake Saimaa and the Finnish nature conservation movement. At present around 300 ringed seals live in Lake Saimaa, and some 60 pups are born annually. Recently animal biometrics, i.e., wild life photo identification have started to offer new ways to monitor animals and to help them to survive. The goal of the SealVision project is to develop automatic image-based indentification of Saimaa ringes seals. Thus, each individual seal could be recognized and tracked in real-time. The approach consists of the detection of a seal in game and wild life camera images, the segmentation of the detected seal, and the identification based on ring patterns on the fur. This is a collaborative project between MVPR and Saimaa Ringed Seal Research Group at University of Eastern Finland (UEF). Contact Prof. Heikki Kälviäinen for futher information.


Computational Psychology of Experience in Human-Computer Interaction

The project aims to study new touch and gesture interaction with novel methodologies. We combine finger biomechanics measurements, eye-movement tracking and experience measurement to get an understanding how users experience new interaction technologies. In experience measurement we utilize a novel quantitative-qualitative experience mapping methods. The data is modeled with Bayesian network, which can reveal relationships between different attributes and parameters via backwards reasoning. This provides novel possibilities to study human experience in interaction. Bayesian networks can be used not only to connect and statistically model single attributes and variables but to construct an entire network that combine all the variables and attributes into one large ensemble. This makes it possible to find and understand completely new connections and form an integrated theory of human interaction experience. This is a collaborative project between MVPR and Psychology of Evolving Media and Technology Research Group (POEM) at University of Helsinki.


Copyright: FIMECC Ltd.

Machine level intelligence, sensing, control, and actuation

The MASCA project is a part of the S-STEP program (Smart technologies for lifecycle performance) by FIMECC (Finnish Metals and Engineering Competence Cluster). The MASCA project focuses on machine vision solutions based on images and other measurements in the applications fields of the S-STEP program. The leading industrial services and the related business models are heavily dependent on reliable, accurate, online information and advance prediction capabilities at the machine and system level. The importance of this information is increasing, requiring also radical improvements in machine level intelligence and independency. While the ICT for heavy industry has been a trend already for decades, the future business models require a revolution also in this area. This revolution is currently known as the emerging industrial internet. The S-STEP programme of FIMECC is positioned in the cross roads of these two significant megatrends: 1) the growing importance of industrial service business, and 2) the remarkable emergence of industrial internet or cyber-physical systems. S-STEP creates the industrial internet technology that enables superior services for the Finnish industry. Contact Prof. Lasse Lensu for futher information.


Traffic sign condition analysis using machine vision

The goal of the TrafficVision project is to support road maintenance, especially in the winter using digital image processing and analysis. The developed methods are used for Finnish Transport Agency (Liikennevirasto, LIVI) to improve their maintenance processes. This kind of machine vision solutions consist of the following tasks: imaging by a camera attached to a road maintenance vehicle, detection of a traffic sign, recognition of the detected traffic sign, and evaluation of the condition of the traffic sign. Contact Prof. Heikki Kälviäinen for futher information.


Imaging and machine vision applications in the wet stage of the pulp and paper production

The PulpVision project focuses on the development of image-based measurement and characterisation methods related to the quality of pulp as a raw material for paper making. PulpVision develops methods to manage the processing of raw materials for fiber-based products. The project develops both entirely new measurements characterizing structures and quality, and aims for transferring research on image-based measurements and off-line methods to on-line measurements and for process quality control.

QVision & EffNet

Image-based measurement methods for quality in pulping and papermaking

The QVision and EffNet projects focuse on image-based measurement methods for quality in papermaking. The projects develop methods such that the functional properties of fiber-based products can be developed, and also managed in production, at a radically improved degree of specificity and scope. The methods are a necessity in the development and production of new high added-value products. The project develops both entirely new measurements characterizing structures and quality, and a procedure for transferring image-based research and off-line methods to on-line measurements and for quality management.


Paper and board printability tests using machine vision

The goal of the Papvision project is to provide novel innovative machine vision and pattern recognition solutions to paper and printing industry. The focus is on novel paper printability assessments which are based on machine vision. Printability tests such as heliotest, picking, and mottling are automatomated using a visual inspection system, called PapVision.


Fusion of visual and digital print quality

The main goal of the DigiQ project is to solve the unrevealed relationship between measurable computational and physical print quality features and subjective print quality experience observed by end-users. Given instrumental vision-based measurements a Visual Quality Index (VQI) is computed based on their observed connection to a human being's subjective quality attributes.


Fusion of computational and visual salience based printed image quality assessment

The goal of the project is to develop a system that is able to categorize a set of unknown images, called Visual Object Categorazation (VOC). Images should be categorized based on objects that images consist of. The system must be unsupervised, called Unsupervised VOC (UVOC), hence it cannot have any prior information about the images, e.g. class labels, training images, etc. From a view point of print quality assessment, the goal is to develop a model for the visual salience of image features and the visual quality assessment, and thus compute a image content based VQI.


Learning real-time analysis and control system for turning


Visual measurement of laser-arc hybrid welding