Patryk Najgebauer

Heuristic Regression Function Estimation Methods for Data Streams with Concept Drift

In this paper the regression function methods based on Parzen kernels are investigated. Both the modeled function and the variance of noise are assumed to be time-varying. The commonly known kernel estimator is extended by adopting two popular tools often applied in concept drifting data stream scenario. The first tool is a sliding window, in which only a constant number of recently received data elements affects the estimator. The second one is the forgetting factor. In this case at each time step past data become less and less important. These heuristic approaches are experimentally compared with the basic mathematically justified estimator and demonstrate similar accuracy.

Content-based image retrieval by dictionary of local feature descriptors

This paper describes a novel method of image key-point descriptor indexing and comparison used to speed up the process of content-based image retrieval as the main advantage of the dictionary-based representation is faster comparison of image descriptors sets in contrast to the standard list representation. The proposed method of descriptor representation allows to avoid initial learning process, and can be adjusted taking into consideration new examples. The presented method sorts and groups components of descriptors in the process of the dictionary creation. The ordered structure of the descriptors dictionary is well suited for quick comparison of images by comparing their dictionaries of descriptors or by comparing individual descriptors with the dictionary. This allows to skip a large part of operations during descriptors comparison between two images. In contrast to the standard dictionary, our method takes into account the standard deviation between the image descriptors. This is due to the fact that almost all descriptors generated for the points indicating the same areas of the image have different descriptors. Estimation of the similarity is based on the determined value of the standard deviation between descriptors. We assume that proposed method can speed up the process of descriptor comparison. It can be used with many solutions which require high-speed operations on the image e.g. robotics, or in software which computes panoramic photography from scrap images and in many others.

Novel method for parasite detection in microscopic samples

This paper describes a novel image retrieval method for parasite detection based on the analysis of digital images captured by the camera from a microscope. In our approach we use several image processing methods to find known parasite shapes. At first, we use an edge detection method with edge representation by vectors. The next step consists in clustering edges fragments by their normal vectors and positions. Then grouped edges fragments are used to perform elliptical or circular shapes fitting as they resemble most parasite forms. This approach is invariant from rotation of parasites eggs or the analyzed sample. It is also invariant to scale of digital images and it is robust to overlapping shapes of parasites eggs thanks to the ability to reconstructing elliptical or other symmetric shapes that represent the eggs of parasites. With this solution we can also reconstruct incomplete shape of parasite egg which can be visible only in some part of the retrieved image.

Properties and structure of fast text search engine in context of semantic image analysis

In the world of computer imaging, we still do not have a good and fast enough method for image searching. This is because science is still not able to imitate fully functions of the human brain. When humans think about images, they do not think about mathematical formulas, matrices, histograms etc. Those mathematical and algorithmic methods are very good for e.g. computer face detection or number plate recognition, but we cannot directly use them for analyzing a whole image and for searching in a set of thousands or even millions of images. On the other hand, computers are able to scan millions of documents, searching for some phrase or even a single word. Fast text search is fully supported by a majority of significant database systems such as Oracle, PostgreSQL or MS SQL Server. The paper presents fast text search engine from another point of view, that is, its application in content based image retrieval.

Extraction of Objects from Images Using Density of Edges as Basis for GrabCut Algorithm

When we think about images, we usually think about that what we can detect by our eyes. It is easy for us, because all of the hard work is already done by our own brain. Human brain extracts from images all information which is currently important. It is not possible to mirror the whole natural process, because now we do not posses enough knowledge about our brain. Nevertheless, a lot of research is devoted to achieve even part of the targets. This is a small steps strategy, so we are not able to do all at once, but we try to test different approaches, combine and develop new digital images processing algorithms. In this paper we present a DOE (Density of Edges) algorithm and its application as a basis for the GrubCut algorithm. We also present the whole preprocessing approach and which algorithms were used. Results of that work will be used and integrated in SIA Semantic Image Analysis project developed by authors.

Representation of Edge Detection Results Based on Graph Theory

This paper describes a concept of image retrieval method based on graph theory, used to speed up the process of edge detection and to represent results in more efficient way. We assume that result representation of edge detection based on graph theory is more efficient than standard map-based representation. Advantages of graph-based representation are direct access to edge nodes of the shape without search and segmentation of edges points as is the case with map-based representations. Another advance is less data consumption, only data for nodes and their connections are needed, what is important in large database applications for good scalability.

Novel method for joining missing line fragments for medical image analysis

We present a method of predictive reconstructing connections between parts of object outlines in images. The method was developed mainly to analyze microscopic medical images but is applicable to other types of images. Examined objects in such images are highly transparent, moreover close objects can overlap each other. Thus, segmentation and separation of such objects can be difficult. Another frequently occurring problem is partial blur due to high image magnification. Large focal length of a microscope dramatically narrows the range of sharp image (depth of field). The method is based on edge detection to extract object contours and represent them in a vector form. The logic behind the presented method refers to the Gestalt Laws describing human perception. The method, according to the law of good continuation and the principle of similarity, evaluates the neighborhood the interrupted contour path, and then tries to determine the most appropriate connection with the other parts of the contour. To assess the similarity of contour parts, the method examines the orientation of the line determined by the gradient of the edge, the characteristics of the edge cross section and the direction of its current course. In order to reduce the amount of data and accelerate the method, fragments of detected outlines are represented by vectors in the form of a graph. Thus, the method has faster access to the information on the course of the edges than in the case of bitmap-based representations.

Novel Image Descriptor Based on Color Spatial Distribution

This paper proposes a new image descriptor based on color spatial distribution for image similarity comparison. It is similar to methods based on HOG and spatial pyramid but in contrast to them operates on colors and color directions instead of oriented gradients. The presented method assumes using two types of descriptors. The first one is used to describe segments of similar color and the second sub-descriptor describes connections between different adjacent segments. By this means we gain the ability to describe image parts in a more complex way as is in the case of the histogram of oriented gradients (HOG) algorithm but more general as is in the case of keypoint-based methods such as SURF or SIFT. Moreover, in comparison to the keypoint-based methods, the proposed descriptor is less memory demanding and needs only a single step of image data processing. Descriptor comparing is more complicated but allows for descriptor ordering and for avoiding some unnecessary comparison operations.

Spatial Keypoint Representation for Visual Object Retrieval

This paper presents a concept of an object pre-classification method based on image keypoints generated by the SURF algorithm. For this purpose, the method uses keypoints histograms for image serialization and next histograms tree representation to speed-up the comparison process. Presented method generates histograms for each image based on localization of generated keypoints. Each histogram contains 72 values computed from keypoints that correspond to sectors that slice the entire image. Sectors divide image in radial direction form center points of objects that are the subject of classification. Generated histograms allow to store information of the object shape and also allow to compare shapes efficiently by determining the deviation between histograms. Moreover, a tree structure generated from a set of image histograms allows to further speed up process of image comparison. In this approach each histogram is added to a tree as a branch. The sub tree is created in a reverse order. The last element of the lowest level stores the entire histogram. Each next upper element is a simplified version of its child. This approach allows to group histograms by their parent node and reduce the number of node comparisons. In case of not matched element, its entire subtree is omitted. The final result is a set of similar images that could be processed by more complex methods.

Improved X-ray Edge Detection Based on Background Extraction Algorithm

Digital X-ray imaging is a source of generous information about health of patient bones. One of major obstacles in computer analysis of digital X-ray images is the presence of bone tissue and soft-tissue areas. It has a negative impact on the quality of bone edge detection or detection of bones area on X-ray images. The main goal is to create an efficient method of edge detection which performs efficiently on properly prepared digital X-ray images. This paper describes a new method of background removal from X-ray images where the background is in the form of soft-tissue. The aim of this is to prepare the image to the next step of processing. We also present a new approach to edge detection of bones on X-ray images. Performance of the proposed method is achieved by eliminating unnecessary areas of the image which are not bone tissue and which are not the main region of interest. Additionally, the presented method of edge detection is partly based on known algorithm named Integral Image and specific edge detection filter, what allow to achieve the desired objectives.