Computer Vision, Graphics, and Image Processing
Rough approximation of shapes in pattern recognition
The paper approaches the nature of a feature recognition process through the description of image features in terms of the rough sets. Since the basic condition for representing images must be satisfied by any recognition result, elementary features are defined as equivalence classes of possible occurrences of specific fragments existing in images. The names of the equivalence classes (defined through specific numbers of objects and numbers of background parts covered by a window) constitute the best lower approximation of window contents (i.e., names of recognized features). The best upper approximation is formed by the best lower approximation, its attributes, and parameters, all referenced to the object fragments situated in the window. The rough approximation of shapes is resistant to accidental changes in the width of contours and lines and to small discontinuities and, in general, to possible positions or changes in shape of the same feature. The rough sets are utilized also on the level of image processing for noiseless image quantization.
References (5)
- Z. Wojcik
An approach to the recognition of contours and line-shaped objects
- Z. Wojcik
A natural approach in image processing and pattern recognition: Rotating neighbourhood technique, self-adapting threshold, segmentation and shape recognition
Pattern Recognit.
(1985)
Cited by (26)
Granular computing, rough entropy and object extraction
2005, Pattern Recognition LettersThe problem of image object extraction in the framework of rough sets and granular computing is addressed. A measure called “rough entropy of image” is defined based on the concept of image granules. Its maximization results in minimization of roughness in both object and background regions; thereby determining the threshold of partitioning. Methods of selecting the appropriate granule size and efficient computation of rough entropy are described.
On links between mathematical morphology and rough sets
2000, Pattern RecognitionBased on the observation that rough sets and mathematical morphology are both using dual operators sharing similar properties, we investigate more closely the links existing between both the domains. We establish the equivalence between some morphological operators and rough sets defined from either a relation, or a pair of dual operators or a neighborhood system. Then we suggest some extensions using morphological thinning and thickening, and using algebraic operators. We propose to define rough functions and fuzzy rough sets using mathematical morphology on functions and fuzzy mathematical morphology.
Evaluations of fractal geometry and invariant moments for shape classification of corn germplasm
1998, Computers and Electronics in AgricultureComputer vision-based techniques were developed and evaluated for classifying different shapes of germplasms (ear of corn). An algorithm was developed to discriminate round-shaped germplasms based on two features, i.e. circularity and dimensional ratio. Two different approaches based on fractal geometry and higher order invariant moments were used for classification of non-round shaped germplasms. In the fractal-based approach, two additional fractal geometry-based features (i.e. fractal-shape factor and fractal perimeter) were developed and used with fractal dimension and aspect ratio to represent the shape features of the germplasms. Classifications rules based on modified Euclidean measures and distance weighted K-nearest neighborhood were used for classifying the germplasms into one of three non-round-shape classes (cylindrical, cylindrical-conical and conical). Though the overall correspondence for classifying non-round germplasms was 60% (based on 80 samples), a maximum correspondence of 80% could be obtained for classifying cylindrical germplasms (based on 18 samples). Neither method could provide similar classification correspondence for cylindrical-conical germplasms. On the other hand, these methods, however, showed a correspondence of 82.5% for classifying non-round corn germplasms into cylindrical and non-cylindrical (conical and cylindrical-conical) shapes.
A method for shift, rotation, and scale invariant pattern recognition using the form and arrangement of pattern-specific features
1992, Pattern RecognitionA new computer based method for visual pattern recognition is described. The process extracts shape features from binary edge images represented in either real-valued or discrete coordinates. Both separate and overlapping features are extracted, maximizing their completeness and continuity, without using templates or mathematical feature descriptions. These features and their arrangement are encoded in a shift, rotation, and scale invariant manner and used to recognize incoming patterns from a set of previously encoded patterns stored in memory. New patterns may be automatically classified and stored in memory, allowing the system to learn and generalize without operator intervention. Testing shows the system to be completely shift, rotation, and scale invariant, and insensitive to local image distortion.
Parallel shape coding on a SIMD architecture
1990, Engineering Applications of Artificial IntelligenceThis paper presents a parallel shape analysis by simultaneous merging of elementary features detected through calculation of the number of objects within neighbouring windows. The parallel algorithm can be embedded on a SIMD mesh architecture. Elementary image features are detected inside partly overlapping windows fixed in the image plane. Each window's contents are processed by a separate processing element (PE). Two neighbouring elementary feature elements are merged by adjacent PEs, and the joined feature chunks are merged in the next step by every 22th PE possessing the same feature. Feature ends are propagated across N = 2n PEs in each nth parallel step toward opposite edges of the mesh array of PEs. The shape coding is completed if all the feature limits (e.g. edges) meet together on one PE. The approach has the property of mapping an image fragment directly into words and phrases.
A rough grammar for a linguistic recognition of image patches
1990, Signal ProcessingThis paper introduces a rough grammar involving explicit operators embedded in the production rules. The grammar is capable of: (a) running the operators on a data-driven or on a data-flow type machine; (b) spreading the hypotheses, goals and constraints from the grammar tree root down to the leaves; (c) coming from local object details (leaves of an execution tree) to a purposeful more global representation (the execution tree root) concurrently avoiding redundant computations; (d) avoiding impasses in a top-down execution tree which need to back up to previous decision points.
The rough grammar is used with a new method of syntactic and semantic object recognition from region patches delimited on one side. The region patches are identified from secants (intersections of scanning lines with image objects. The nearest neighbor clustering combined with ordering coordinates of secant ends appends an end of a current secant to a cluster of a patch edge in case the end is closest to the most recent element of the patch edge and in case the end satisfies an assumed order of coordinates. The consideration of the following factors: secant beginnings, secant endings, non-ascending order of secant ends and non-descending order of secant ends permits the region patches to be clustered into basic types. Details of the derivation of patches are presented to show how demands can be propagated from higher-level rough-grammar rules down to the clustering operators.
To identify a corner patch, two continuous edge patches belonging to different clusters of region patches are merged if the continuous patches are of efficient lengths and if they are close enough to meet each other. Polygons are identified by merging corner patches or three-sided region patches of different types. Examples of identification of simple objects are presented. The process of identification of different objects can be performed in parallel using separate processors.
Dieser Beitrag stellt eine Methode für die Erkennung und Rekonstuktion von Objektregionen aus Bildausschnitten vor. Die Signalverarbeitung beginnt damit, daβ alle Bildobjekte in Sekanten aufgespaltet werden. Im nächsten Schritt werden Regionen aus den Ausschnitten rekonstruiert. Die Untersuchung auf Häufungsgebiete mit Hilfe der Nächste-Nachbar-Regel ermöglicht es, das Ende einer Sekante an eine Ballung von Kantenregionen auzuhängen, falls das fragliche Ende dem zuletzt angefügten Element der Kantenregion am nächsten liegt und falls das Ende eine vorgegebene Rangfolge der Koordinaten einhält. Das neue, nächstgelegene Sekantenende wird von der nächsten Zeile des Bildes aus angewählt. Die Überlagerung der vier Gröβen Sekantenanfang, Sekantenende, nichtaufsteigende Reihenfolge der Sekantenenden und nichtabsteigende Reihenfolge der Sekantenenden erlaubt es, Kantenregionen in vier Grundtypen einzuteilen. Die Kantenregionen werden dann in kontinuierliche Gebietsstreifen zusammengefaβt: falls ein Kandidat ein vorgegebenes Kriterium nicht befriedgt, wird das zuletzt hinzugefügte Element des Häufungsgebieter als das letzte Element eines kontinuierlichen Kantenbereichs betrachtet. Die so entdeckten kontinuierlichen Gebietsstreifen sind tatsächlich Gebiete, die von einer Seite her durch die Kanten begrenzt werden.
Der nächste Schritt besteht darin, jeden dieser Gebietsstreifen präziser zu begrenzen; dies geschieht durch einen zweiten Rand, der mit dem ersten eine Ecke bildet, oder aber mit dem gleichen, jedoch nunmehr kurvenförmigg verlaufenden Rand.
Um eine Gebietsecke zu rekonstruieren, werden zwei kontinuierliche Gebietsstreifen zusammengefaβt, falls diese von ausreichender Länge sind und nahe genug beieinander liegen. Polygone werden zusammengefaβt, indem Gebiete mit Ecken oder Gebiete verschiedener Art mit jeweils drei Seiten vereinigt werden. Für die Rekonstruktion polygonförmiger Objekte werden Beispiele angegeben. Die Rekonstruktion auseinanderliegender Objekte kann in Parallelverarbeitung unter Einsatz separater Prozessoren erfolgen.
Ce papier introduit une grammaire approximative dont les régles de production contiennent des opérateurs explicites. la grammaire est capable: (a) d'activer les opérateurs sur une machine dirigée par les données ou à flot de données; (b) d'étendre les hypothèses, buts et contraintes à partir de la racine de l'arbre de la grammaire jusqu'aux feuilles; (c) de passer des détails d'objets locaux (les feuilles de l'arbre d'exécution) à une représentation déterminée plus globale (la racine de l'arbre d'exécution) en évitant les calculs redondants; (d) d'éviter des impasses dans l'arbre d'exécution en mode descendant qui nécessitent la copie des points de décision précédents.
La grammaire approximative est utilisée dans une nouvelle méthode de reconnaissance syntaxique et sémantique d'objets à partir d'éléments de régions bornés d'un seul côte. Les éléments de région sont identifiés à partir des sécantes (les intersections des lignes de balayage et de l'image des objets). Le groupement par la règle du voisin le plus proche est combiné avec la mise en ordre des coordonnées des fins de sécantes: la fin de la sécante courante est assignée à un élément borné si la fin est la plus proche du point le plus récent de l'élément borné et si la fin vérifie un ordre prédéterminé pour les coordonnées. Les éléments de régions seront classés suivant des types de base en considérant les facteurs suivants: début et fin de sécants, ordre non-ascendant et non-descendant des fins de sécantes. Des détails sur l'obtention des éléments seront présentés pour montrer comment des demandes peuvent être propagées à partir des règles de haut-niveaux de la grammaire jusqu'aux opérateurs de groupement.
Pour identifier un élément anguleuz, deux éléments bornés continus appartenant à différents groupes d'éléments de région sont fusionnés si les éléments continus sont de longueur suffisante et si ils sont assez proches pour se rencontrer. Des polygones sont identifiés par fusion d'éléments anguleux ou de trois éléments de région bornés d'un côté, de types différents. Des exemples d'identification d'objets simples sont présentés. Le processus d'identification des différents objets peut être réalisé en parallèle en utilisant plusieurs processeurs séparés.