Zoltan Vamossy

Sketch4match — Content-based image retrieval system using sketches

The content based image retrieval (CBIR) is one of the most popular, rising research areas of the digital image processing. Most of the available image search tools, such as Google Images and Yahoo! Image search, are based on textual annotation of images. In these tools, images are manually annotated with keywords and then retrieved using text-based search methods. The performances of these systems are not satisfactory. The goal of CBIR is to extract visual content of an image automatically, like color, texture, or shape. This paper aims to introduce the problems and challenges concerned with the design and the creation of CBIR systems, which is based on a free hand sketch (Sketch based image retrieval — SBIR). With the help of the existing methods, describe a possible solution how to design and implement a task spesific descriptor, which can handle the informational gap between a sketch and a colored image, making an opportunity for the efficient search hereby. The used descriptor is constructed after such special sequence of preprocessing steps that the transformed full color image and the sketch can be compared. We have studied EHD, HOG and SIFT. Experimental results on two sample databases showed good results. Overall, the results show that the sketch based system allows users an intuitive access to search-tools. The SBIR technology can be used in several applications such as digital libraries, crime prevention, photo sharing sites. Such a system has great value in apprehending suspects and indentifying victims in forensics and law enforcement. A possible application is matching a forensic sketch to a gallery of mug shot images. The area of retrieve images based on the visual content of the query picture intensified recently, which demands on the quite wide methodology spectrum on the area of the image processing.

Fuzzy based obstacle avoidance for mobil robots with Kinect sensor

This article describes an improved version of a fuzzy-based navigational system and compares the developed system with the Caversan [11] navigational method that was used as a basis of our approach. Our system (which uses a Mamdani-type fuzzy controller that has sixteen rules, three inputs and one output) has better algorithms for both the path planning and for the avoiding of obstacles. This is because our approach takes into consideration both the point-like obstacles in the front, the close walls, and the obstacles on the left as well; these are the aspects where the reference system makes a false assumption and the robot continues going straight forward. Our new system does a sharp turn if it detects an obstacle that is close, and goes on an almost ideal track if the obstacles are further away. The developed system was tested both in the MATLAB fuzzy simulator and using an autonomously moving robotic car as well; and it was determined that the robot is capable of navigation in real time while it avoids the obstacles. Since the evaluation of the environment is done using Kinect and infra-based distance-detection sensors, we introduced a new fuzzyfication method. Because of this, the parameters of the original method had to be adjusted to match the sensors field of view and area of detection (this had to be done separately for the distances on the sides and in the front as well).

Differences between Kinect and structured lighting sensor in robot navigation

Nowadays, autonomous navigation is getting more and more attention. The mobile robots need to have map of the close environment. The paper presents how can collect information about the environment for robot navigation. The sensors and the navigation system are installed on a mobile equipment platform. After this the article shows the different measurement systems, the generating virtual depth map and the navigation algorithm. At the end the Kinect sensor is compared with a sensor based on structured light.

Parallel biomedical image processing with GPGPUs in cancer research

The main aim of this work is to show, how GPGPUs can facilitate certain type of image processing methods. The software used in this paper is used to detect special tissue part, the nuclei on (HE - hematoxilin eosin) stained colon tissue sample images. Since pathologists are working with large number of high resolution images - thus require significant storage space -, one feasible way to achieve reasonable processing time is the usage of GPGPUs. The CUDA software development kit was used to develop processing algorithms to NVIDIA type GPUs. Our work focuses on how to achieve better performance with coalesced global memory access when working with three-channel RGB tissue images, and how to use the on-die shared memory efficiently.

Virtual Hand - Hand Gesture Recognition System

The goal of this project to create a hand gesture recognition application, which is useful in teaching hearing-impaired or deaf people. The hardware component of the system is a stereo camera, which made from two webcams. The hand detection based on a preprocess part, that is a HSV color conversation with an intensity based threshold filter, and a main process part, which is a thinning skeletonization algorithm. The solution of the recognition process is solved by a neural network. The input layer get vectors, that are created from the main feature points of the detected hand, and the output layer represents the parameters of a three dimensional hand model.

Traffic monitoring with computer vision

Nowadays roads and streets are getting overcrowded, especially in bigger cities. Hence the main goal of our project is to build a traffic monitoring system which is able to detect the movement of cars and to track and count the different vehicles by analyzing a camera picture with the help of computer vision. The real-time process (15–30 fps) of the video stream works at daylight. The traffic monitoring includes the followings: it accepts the video from a file or a camera, marks the moving vehicles with rectangles, and counts them. The system consists of three subsystems, these are the: Video Subsystem, Motion Detector Subsystem, Display/Control Subsystem. To maximize the speed of the program, each subsystem runs on different threads. To accept the visual information DirectShow is used, and the image processing is partially done with the DirectCV wrapper [6] for OpenCV. For motion detection, the system uses a Gaussian Mixture Model with background segmentation. There is an opportunity to mask the monitoring area also. With the mask it is possible to filter out for example the sidewalks and the opposite traffic direction so that our system avoids these parts of the image. The mask area can be easily built by placing a sequence of points which are the vertexes of a polygon. To create a polygon from the sequence of points a hybrid algorithm is applied which combines the 2opt heuristic method and a genetic algorithm. Connected component labeling algorithm is used for object detection, which utilizes a decision tree and Union-Find data structure in order to achieve the best performance. Recently we are testing a shadow removal algorithm based on edge detection in order to detect the objects in traffic scenes more accurately.

Map Building and Localization of a Robot Using Omnidirectional Image Sequences

The paper describes a map building module, where the image sequences of the omnidirectional camera are transformed into virtual top-view ones and melted into the global dynamic map. After learning the environment from training images, a current image is compared to the training set by appearance-based matching. Appropriate classification strategies yield an estimate of the robots current position.

Preparing initial population of genetic algorithm for region growing parameter optimization

The processing of microscopic tissue images is nowadays done more and more using special immunodiagnostic-evaluation software products. Often to evaluate the samples, the first step is determining the number and location of cell nuclei. To do this, one of the most promising methods is the region growing, but this algorithm is very sensitive to the appropriate setting of different parameters. Due to the large number of parameters and due to the big set of possible values setting those parameters manually is a quite hard task, so we developed a genetic algorithm to optimize these values. The first step of the development is the statistical analysis of the parameters, and the determination of the important features, to extract valuable information for a to-be-implemented genetic algorithm that will perform the optimization.

Thermal image fusion

The goal of the project is to compile a single picture from a series of shots made by a thermal camera. The reason for that is on the assembled picture it is much easier to examine the workspace important for the user. Because the initial series is shot by the same kind of camera, the traditional four phases can be used for the fusion: detection of feature points; matching of feature points; transformation model determination and creation; and finally the transformed sensed image inserted into the reference image using averaging. However, during the matching and the creation of the transformation model we need to take into account that these pictures are shot by non-traditional cameras, so their attributes would be somewhat different from the usual. The system is semiautomatic, and its efficiency is tested both indoors and outdoors.

Mobile robot navigation in unknown environment using structured light

Our paper represents how it is possible to control a mobile robot through computer vision. In the paper the issues of emergency response (automatic obstacle avoidance) and autonomous control is mentioned as well. For perception, the robot uses a laser diode and a webcam. The dot shaped light of the laser diode is transformed to a line by an optics. On the perceived image the spatial position of objects can be obtained from the reference position of the projected line and the actual position of the laser line. To specify the possible obstacles on the planned route of the vehicle, the depth map which can be obtained by the triangular method can be used. The completed software runs on a vehicle mounted computer and communicates with the car through serial port.