The partially supervised Compositional Zero-Shot Learning (pCZSL) recognizes new compositions of states and objects, where for every image in the training set either the state or the object annotation is available. In pCZSL, features of a state vary depending on the object in the composition (e.g. the features of state ripe are different for ripe banana and ripe apple). Understanding the variation in features across scales of objects is also a key challenge. In the proposed architecture, a swin transformer based Hierarchical Feature Extractor (HFE) captures the large range of semantic interactions between state and object features. The Discriminative Context Aggregation module utilizes features from the intermediate layers of the HFE to understand the features of object at their corresponding scales. To leverage the partially labeled data in pCZSL, we pass strongly and weakly augmented versions of the input image to the proposed architecture. The predicted class probabilities for strongly and weakly augmented images are encouraged to be similar, minimizing a distribution alignment loss. This loss incorporates class specific re-weighting approach to alleviate the effect of data imbalance for pCZSL. Extensive experiments on three benchmark datasets demonstrate the superiority of the proposed approach.

This paper proposes panoptic segmentation to identify regions of interest (RoIs) in medical images where intensity profiles of RoIs have significant overlap and inconsistent across different instances of same type of RoI. Inspired by the proposal of unified architecture and unified segmentation model for multi-task segmentation, namely semantic, instance and panoptic segmentation, we propose a custom panoptic segmentation scheme for the non-small cell lung cancer (NSCLC) dataset. A typical challenge of such dataset is to differentiate image segments representing primary and secondary focus of tumours having non-homogeneous and overlapping intensity profile. In the proposed architecture, outputs of image feature extractor for each 2D CT slice images and the corresponding text based description of the ground truth RoIs (e.g. photo of left lung) are mapped to a transformer-based decoder to generate segmented output. During inference, image along with a task query (e.g. panoptic or instance segmentation) generates test segmentation map. Based on a conservative performance metric introduced in this paper, we have shown that segmentation by the proposed model is better than its close competitors.

Compositional Zero-shot Learning (CZSL) attempts to recognise images of new compositions of states and objects when images of only a subset of state-object compositions are available as training data. An example of CZSL is to recognise images of peeled apple by a model when it is trained using images of peeled orange , ripe apple and ripe orange . There are two major challenges in solving CZSL. First, the visual features of a state vary depending on the context of a state-object composition. For example state like ripe produces distinct visual properties in the compositions ripe orange and ripe banana . Hence, understanding the context dependency of state features is a necessary requirement to solve CZSL. Second, the extent of association between the features of a state and an object varies significantly in different images of same composition. For example, in different images of peeled oranges , the oranges may be peeled to different extents. As a consequence, the visual features of images of the class peeled orange may vary. Hence, there exists a significant amount of intra-class variability among the visual features of different images of a composition. Existing approaches merely look for the existence or absence of features of particular state or object in a composition. Our approach not only looks for the existence of a particular state features or object features but also the extent of association of state features and object features to better tackle the intra-class variability in visual features of compositional images. The proposed architecture is constructed using a novel Knowledge Guided Transformer . The transformer-based framework is utilised for processing larger context dependency between the state and object. Extensive experiments on C-GQA, MIT-States and UT-Zappos50k datasets demonstrate the superiority of the proposed approach in comparison with the state-of-the-art in both open-world and closed-world CZSL settings.

The purpose of Compositional Zero-Shot Learning (CZSL) is to recognise unseen (or new) compositions of known objects (e.g. apple, banana) and its states (e.g. peeled, ripe). The CZSL poses a challenging problem as the visual features of objects and states are often difficult to isolate from their composition in images. Also, the features of a state may have wide variation across different compositions. For example, the state sliced presents different visual features in compositions, sliced potato and sliced bread. Hence, for solving CZSL problem, disentanglement of state and object features is the key challenge that we attempt to address in this paper by introducing Bi-modal Compositional Network (BiComNet). The proposed BiComNet consists of two distinct branches for state and object recognition, which helps in better disentanglement of the state and object features. We also propose a novel loss component to better facilitate the disentanglement of the state and object features. Finally, the proposed BiComNet integrates two different modalities of input data, the visual and word features, to better recognise unseen state-object compositions. Extensive experiments are carried out on the benchmark datasets MIT-States & UT-Zappos50K. Our proposed algorithm outperforms competitive methods in most count.

Human pose estimation in 3D from monocular images is a challenging inverse problem due to ambiguity in lifting 2D projection to 3D space. In this article we have made three contributions in order to solve 3D pose estimation. First, a new DNN architecture is proposed to generate multiple feasible 3D pose hypotheses from a given image. Second, we generate weights for the proposed hypotheses using ordinal supervision. These weights are used to predict the final 3D pose from the generated hypotheses. Finally, we report a new regularizer to enforce that the predicted skeleton is consistent with the restriction of anthropomorphic constraints. We compare the results of our algorithm with other state-of-the art approaches on the Human 3. 6m benchmark dataset. Our algorithm reports competitive results.

The problem of computer‐aided detection of architectural distortion (AD) in a digitised mammogram has been attempted in this manuscript. In examining a mammogram, the decision regarding a particular region of interest (RoI) is dependent on the appearance of the surrounding regions. However, in existing methods to detect AD the inference about an RoI is dependent on the appearance of this RoI alone. In addition, multiple radiologists infer the same mammogram in coming to a final decision about the mammogram. Contrary to popular ensemble classifiers like Adaboost and Random Forest, the authors propose an ensemble based method (imitating multiple radiologists by classifiers) for detecting AD such that the decision on a test RoI is dependent on the decisions of the surrounding RoIs in the proposed ensemble classifier. The proposed context‐based ensemble classifier has been validated on two mammographic databases. The proposal shows promising results in both the databases.

Rheumatologists determine treatment plan based on the inflammation of knee joints affected by arthritis. Extraction of the inflamed region or hotspot from the knee thermogram is the prerequisite for grading of inflammation and classification of different arthritis. In this paper, we propose an automatic method for extracting the inflamed region from the knee thermograms. We propose an ensemble technique to arrive at a consensus segmentation of the hotspot region. We have used variation of information based information theoretic approach to generate consensus segmentation. The fusion of multiple segmentation maps is achieved using local search based greedy iterated conditional modes algorithm to obtain final segmentation result. Experiments show that our proposal scores significantly better in detecting hotspots in more than 50 inflammatory knee thermograms.

We propose a shape-based solution for speckle removal from ultrasound images. The method is operable in both low contrast and high contrast imaging scenarios. The approach introduces shape information alongside structural information in the speckle removing filter. By iteratively minimizing a shape fidelity penalty to reduce speckle, the proposed filter facilitates superior segmentation of blood vessels. The effectiveness of the proposed method is established through experimentation on the ultrasound images of human blood vessels. The results show at more than 7% improvement in PSNR values compared to other state-of-the-art approaches.

We propose a fast and accurate solution to speckle reduction targeted specifically at optical coherence tomography images. The proposed speckle removing filter is designed using a novel potential function based on the gradient of the local variance of intensity. After filtering, the spatially neighboring pixels with close values of intensities converge to uniform gray values, while the edges remain intact. This filtering process results in removal of speckle without destroying the edges of the desired object. The proposed filter also prevents the generation of any false edges. Detailed experimental analysis shows at least 1-dB improvement in the peak signal-to-noise ratio for spectral domain optical coherence tomography images. The method also shows superior edge preservation, contrast, and speed compared to the state of the art in speckle removing filters.

For synthesis of realistic facial expressions displaying emotions, we need an efficient representation of pure (e.g., surprise) as well as mixed (e.g., happily surprised) emotional expressions. In this paper, we train an expression map (XM) that efficiently represents the emotional expressions. We propose an algorithm that utilizes the XM to synthesize emotional expressions, tailor-made for the facial structure of the target person. The proposed method can also control the proportions of different basic emotional expressions of those, when mixed together, to generate realistic emotional facial expressions. Unlike many existing methods, our expression synthesis model requires only one expression-neutral face image of the target person. Both qualitative and quantitative tests on four data sets show promising results. On average, we have achieved 92.4% correct validation of the expressions synthesized by our method. We also show that for both basic and mixed emotional expressions, our method generates finer expression details compared to existing state-of-the-art works.

Classification using cross-datasets (where a classifier trained using annotated image set A is used to test similar images of set B due to lack of training images in B) is important for many classification problems especially in biomedical imaging. We propose a discriminative autoencoder, useful for addressing the challenge of classification using cross-datasets. Our autoencoder learns an encoder and decoder such that the distances between the representations of the same class is minimized whereas the distances between the representations of different classes are maximized. We derive a fast algorithm to solve the aforementioned problem using the Augmented Lagrangian Alternating Directions Method of Multipliers (ADMM) approach. ADMM is a faster alternative to back-propagation which is used in standard autoencoders. The proposed method outperforms state-of-the-art representation learning tools in terms of classification results in breast cancer related histopathological image set MITOS and AMIDA and some of the benchmark image datasets.

Proportions of different phases (phase fraction) in the microstructures determine the quality of dual phase (DP) steel. So, calculation of phase fraction in the microstructures of steel samples is important for quality assurance. Manual calculation of phase fraction involves Le Pera etching of steel which is time consuming and dependent on operator efficiency. Calculation of phase fraction from Le Pera etched samples requires cumbersome manual observations. Nital etching is a faster alternative to Le Pera etching. However, due to lack of visually discriminative information, different phases cannot be identified manually from nital images. We propose a novel method for automatic calculation of phase fractions in steel microstructures from nital images using machine learning techniques. We show that regional contour patterns and local entropy (which cannot be evaluated manually) of regions of nital images are related to the formation process of the phases. We design a method that automatically evaluates regional contour patterns and local entropy from nital images of DP steel. Subsequently, we construct a random forest classifier that uses regional contour patterns and local entropy as features for classification of different phases. Our method is ∼150 times faster than manual classification. Experiments show close to 90% accuracy in classification.

We propose an improved random forest classifier that performs classification with minimum number of trees. The proposed method iteratively removes some unimportant features. Based on the number of important and unimportant features, we formulate a novel theoretical upper limit on the number of trees to be added to the forest to ensure improvement in classification accuracy. Our algorithm converges with a reduced but important set of features. We prove that further addition of trees or further reduction of features does not improve classification performance. The efficacy of the proposed approach is demonstrated through experiments on benchmark datasets. We further use the proposed classifier to detect mitotic nuclei in the histopathological datasets of breast tissues. We also apply our method on the industrial dataset of dual phase steel microstructures to classify different phases. Results of our method on different datasets show significant reduction in average classification error compared to a number of competing methods.

A three‐dimensional (3D) representation of the linear structures in a breast has been constructed by utilising information of the ridges in the craniocaudal (CC) and the mediolateral oblique (MLO) mammographic views of the breast. Blood vessels and ducts appear as ridges in the mammogram. The position and shape of the ridges in the breast are an indicator of malignancy. In the first stage of the proposed 3D reconstruction problem, an algorithm has been developed to find out which linear structure in the CC view and a linear structure in the MLO view correspond to the same 3D structure in the breast. The 3D view of the linear structures is constructed from the abovementioned correspondences. The positional error (per unit length of the reconstructed linear structure) between the manual reconstruction and the algorithmic reconstruction of a linear structure turns out to be better by 20% than a competing approach.

Grading of cancer offers insight to the occurrence and progress of the disease. The course of treatment is planned depending on the grade of cancer. Segmentation of the glandular structure of tissue is a prerequisite for grading of colon, prostate and breast cancers. Manual segmentation method is time-consuming and suffers from the curse of observer bias. We propose an automated solution for gland segmentation from hematoxylin & eosin (H&E) stained histology images. Our method relies on the biological cue rather than gland specific signatures that may vary across the slides. We construct a novel informative morphological scale space for gland segmentation. The scale space uses the entropy of the connected components in a novel manner to prevent over segmentation of objects. Our solution is fast, accurate and applicable in a clinical setup. Experiments show an average F1 score of 0.68 for 85 histology images in 20x magnification. We obtain ~ 30% improvement in F1 score compared to the area morphological scale space method.

In this paper, we propose a novel scheme for automatic recognition of facial expressions captured from both fronto-parallel and non-fronto-parallel cameras i.e., multi-view facial expressions (MVFE). The proposed scheme introduce a Local Saliency-inspired Binary Pattern (LSiBP) feature to recognize MVFE. First view-specific approximated saliency likelihood map (ASLM) is derived during training of our model. ASLM is determined from the 2D structure tensor representation of faces. The distribution of saliency likelihoods of pixels along with pixel intensities are analyzed in extracting LSiBP features from a face. Such LSiBP features are utilized for training and testing of view-specific SVM classifiers. Extensive experiments are performed on datasets of both posed and unconstrained spontaneous expressions of MVFE. Our scheme outperforms state-of-the-arts by at least 1% to recognize MVFE.

We propose an action recognition technique using the 3D skeleton model of human without compromising the identity of the person. The skeleton model is defined as a set of 3D joint (e.g. knee or hip joint) coordinates obtained from the Kinect. The low frequency sensor noise in estimating the joint coordinates is removed after modeling the covariance matrix of the joint coordinates as a function of variance of individual joint coordinates. We determine a range for the threshold of this covariance matrix to detect active joints defining an action. Since, a sparse set of active joint coordinates is enough to represent an action, we map these coordinates to lower dimensional linear manifold before training using an SVM classifier. The recognition rate using our proposed approach outperforms competing approaches by at least 2%.

Histopathological grading of cancer not only offers an insight to the patients' prognosis but also helps in making individual treatment plans. Mitosis counts in histopathological slides play a crucial role for invasive breast cancer grading using the Nottingham grading system. Pathologists perform this grading by manual examinations of a few thousand images for each patient. Hence, finding the mitotic figures from these images is a tedious job and also prone to observer variability due to variations in the appearances of the mitotic cells. We propose a fast and accurate approach for automatic mitosis detection from histopathological images. We employ area morphological scale space for cell segmentation. The scale space is constructed in a novel manner by restricting the scales with the maximization of relative-entropy between the cells and the background. This results in precise cell segmentation. The segmented cells are classified in mitotic and non-mitotic category using the random forest classifier. Experiments show at least 12% improvement in F1 score on more than 450 histopathological images at 40× magnification.

A global solution to the enhancement and segmentation of ultrasound images is proposed that is operable in both low contrast and high contrast imaging scenarios. The solution is based on two optimization processes: one that minimizes error with respect to the original image while minimizing the number of edge contours (above the number expected by the known topology), and the second that maximizes edge fidelity. The maximization of edge fidelity is achieved by way of connected filters that operate on connected components of a threshold-decomposed image. To test the algorithm, an application in the ultrasound imaging of human blood vessels is explored. The results show significant improvements (8X to 50X) in a vesselness measure of the segmented vessels as compared to that yielded by a traditional speckle reduction technique and to a diffusion-based technique.

In this paper, we propose a graph theoretic approach for recognizing interactions between two human performers present in a video clip. We watch primarily the human poses of each performer and derive descriptors that capture the motion patterns of the poses. From an initial dictionary of poses (visual words), we extract key poses (or key words) by ranking the poses on the centrality measure of graph connectivity. We argue that the key poses are graph nodes which share a close semantic relationship (in terms of some suitable edge weight function) with all other pose nodes and hence are said to be the central part of the graph. We apply the same centrality measure on all possible combinations of the key poses of the two performers to select the set of 'key pose doublets' that best represent the corresponding action. The results on standard interaction recognition dataset show the robustness of our approach when compared to the present state of the art method.

In this paper, we propose a graph theoretic technique for recognizing human actions at a distance in a video by modeling the visual senses associated with poses. The proposed methodology follows a bag-of-word approach that starts with a large vocabulary of poses (visual words) and derives a refined and compact codebook of key poses using centrality measure of graph connectivity. We introduce a “meaningful” threshold on centrality measure that selects key poses for each action type. Our contribution includes a novel pose descriptor based on histogram of oriented optical flow evaluated in a hierarchical fashion on a video frame. This pose descriptor combines both pose information and motion pattern of the human performer into a multidimensional feature vector. We evaluate our methodology on four standard activity-recognition datasets demonstrating the superiority of our method over the state-of-the-art.

We develop an adaptive active contour tracing algorithm for extraction of spinal cord from MRI that is fully automatic, unlike existing approaches that need manually chosen seeds. We can accurately extract the target spinal cord and construct the volume of interest to provide visual guidance for strategic rehabilitation surgery planning.

We introduce a theoretical framework for implicit evolution of an open ended curve in a two-dimensional image plane. This approach is particularly suitable for identifying thin filamentous structures present in 2D images. The open ended curve is represented as the of the level set of a function (called the*) defined on the curve. The iterative evolution of the curvature map is guided by a diffusion equation and constrained by the imaging force, such as the image intensity gradient. The centerline of the evolved curvature map provides the position of the curve in subsequent iterations. We have tested this new model on both synthetic and real images that contain structures including rivers/roads/arteries. Nine experiments show that our model is successful in identifying complex topological filaments with a low 9% RMSE pixel error, and that the technique withstands effects of shape irregularities such as kinks, bending, circularity and inconsistent edges.

In this paper, a generalized statistical tool is introduced to estimate key frames in a video sequence. The tool works based on the inter-relationship between different features of image frames in a video. The image feature vectors are plotted in feature space as points and a randomness measure is determined from the distribution of these points. The randomness measure of the feature vectors is defined with respect to simulated random point patterns and expressed as a probability value of a frame being a key frame. Since, depending on the video content more than one inter-relationship of features can be used to determine a single key frame, different probability values are derived to support a frame as a key frame. To integrate these probability values a combiner model is designed to uniquely decide the status of a key frame. The combiner model is based on the Dempster-Shafer theory of evidence. To demonstrate the idea, randomness measures, and consequently the probabilities of a frame to be a key frame, are obtained separately from spatial domain and frequency domain features. The combined probability value enhances the confidence in selecting a frame as a key frame. The result is tested on a number of standard video sequences and it outperforms the related approach

We present a novel approach to constraining the evolution of active contours used in image analysis. The proposed approach constrains the final curve obtained at convergence of curve evolution to be related to the initial curve from which evolution begins through an element of a desired Lie group of plane transformations. Constraining curve evolution in such a way is important in numerous tracking applications where the contour being tracked in a certain frame is known to be related to the contour in the previous frame through a geometric transformation such as translation, rotation, or affine transformation, for example. It is also of importance in segmentation applications where the region to be segmented is known up to a geometric transformation. Our approach is based on suitably modifying the Euler-Lagrange descent equations by using the correspondence between Lie groups of plane actions and their Lie algebras of infinitesimal generators, and thereby ensures that curve evolution takes place on an orbit of the chosen transformation group while remaining a descent equation of the original functional. The main advantage of our approach is that it does not necessitate any knowledge of nor any modification to the original curve functional and is extremely straightforward to implement. Our approach therefore stands in sharp contrast to other approaches where the curve functional is modified by the addition of geometric penalty terms. We illustrate our algorithm on numerous real and synthetic examples.

We propose a cell detection and tracking solution using image-level sets computed via threshold decomposition. In contrast to existing methods where manual initialization is required to track individual cells, the proposed approach can automatically identify and track multiple cells by exploiting the shape and intensity characteristics of the cells. The capture of the cell boundary is considered as an evolution of a closed curve that maximizes image gradient along the curve enclosing a homogeneous region. An energy functional dependent upon the gradient magnitude along the cell boundary, the region homogeneity within the cell boundary and the spatial overlap of the detected cells is minimized using a variational approach. For tracking between frames, this energy functional is modified considering the spatial and shape consistency of a cell as it moves in the video sequence. The integrated energy functional complements shape-based segmentation with a spatial consistency based tracking technique. We demonstrate that an acceptable, expedient solution of the energy functional is possible through a search of the image-level lines: boundaries of connected components within the level sets obtained by threshold decomposition. The level set analysis can also capture multiple cells in a single frame rather than iteratively computing a single active contour for each individual cell. Results of cell detection using the energy functional approach and the level set approach are presented along with the associated processing time. Results of successful tracking of rolling leukocytes from a number of digital video sequences are reported and compared with the results from a correlation tracking scheme.

Most of the existing watermarking processes become vulnerable when the attacker knows the watermark insertion algorithm. This paper presents an invisible spatial domain watermark insertion algorithm for which we show that the watermark can be recovered, even if the attacker tries to manipulate the watermark with the knowledge of the watermarking process. The process incorporates buyer specific watermarks within a single multimedia object, and the same multimedia object has different watermarks that differ from owner to owner. Therefore recovery of this watermark not only authenticates the particular owner of the multimedia object but also could be used to identify the buyer involved in the forging process. This is achieved after spatially dividing the multimedia signal randomly into a set of disjoint subsets (referred to as the image key) and then manipulating the intensity of these subsets differently depending on a buyer specific key. These buyer specific keys are generated using a secret permutation of error correcting codes so that exact keys are not known even with the knowledge of the error correcting scheme. During recovery process a manipulated buyer key (due to attack) is extracted from the knowledge of the image key. The recovered buyer key is matched with the exact buyer key in the database utilizing the principles of error correction. The survival of the watermark is demonstrated for a wide range of transformations and forging attempts on multimedia objects both in spatial and frequency domains. We have shown that quantitatively our watermarking survives rewatermarking attack using the knowledge of the watermarking process more efficiently compared to a spread spectrum based technique. The efficacy of the process increases in scenarios in which there exist fewer numbers of buyer keys for a specific multimedia object. We have also shown that a minor variation of the watermark insertion process can survive a "Stirmark" attack. By making the image key and the intensity manipulation proms specific for a buyer and with proper selection of error correcting codes, certain categories of collusion attacks can also be precluded.

In this paper, we propose a buyer authentication scheme that ensures copyright protection of multimedia objects. We show that the authentication mark is robust and recoverable in spite of a number of intentional attacks to destroy the authentication mark. The authentication mark is introduced in the wavelet space. The wavelet coefficients of the multimedia signal are randomly grouped into a number of disjoint subsets of the wavelet coefficients. However, the manipulation of the wavelet coefficient is such that the signal quality is always within the just noticeable distortion (jnd) limit. The buyer specific key survives even though we assume that a potential forger knows the proposed authentication scheme and this knowledge can be used as an attack.

We propose an image segmentation model using an agglomerative clustering technique. The clustering is performed within a feature matrix where intensity and boundary relations are defined between neighboring segments. The iterative segment agglomeration process satisfies a cluster aggregation property and does not utilize a priori knowledge of the number of clusters present in the image. The performance of the algorithm is demonstrated on a number of color images and compared with a similar algorithm.

The problem of cloud tracking within a sequence of geo-stationary satellite images has direct relevance to the analysis of cloud life cycles and to the detection of cloud motion vectors (CMVs). The proposed approach first identifies a homogeneous consistent cloud mass for tracking and then establishes motion correspondence within an image sequence. In contrast to the crosscorrelation based approach as adopted in automatic CMV detection analysis, a scale space classifier is designed to detect cloud mass in the source image taken at time t and the destination image at time t+/spl delta/t. Boundaries of the extracted cloud segments are matched by computing a correspondence between high curvature points. This shape based method is capable of tracking in the cases of rotation, scaling, and shearing, while the correlation technique is limited to translational motion. The final tracking results provide motion magnitude and direction for each contour point, allowing reliable estimation of meteorological events and wind velocities aloft. With comparable computational expense, the scale space classification technique exceeds the performance of the traditional correlation-based approach in terms of reduced localization error and false matches.

We provide an automated method to repair broken, occluded oriented image textures. Our approach is based on partial differential equations (PDEs) and AM-FM image modeling. Reconstruction of the texture occurs via simultaneous PDE-generated diffusion and reaction. In the diffusion process, the image is adaptively smoothed, preserving important boundaries and features. The reaction process produces the reconstructed textural information in the occluded image regions. Gabor (1946) filters are designed and used in the reaction process using an AM-FM dominant component analysis. An AM-FM model of the texture image is constructed, making it possible to localize the reaction filters spatio-spectrally. In contrast to previous disocclusion techniques that depend on interpolation, on continuity of the connected components within the image level sets, or on texture estimation, the reaction-diffusion process proposed here yields a seamless transition between the recreated region and the unoccluded image regions. Using AM-FM dominant component analysis, we avoid the ad hoc parameter selection typified with other reaction-diffusion approaches. As a useful example, we focus on the repair of broken, occluded fingerprints. We also treat several exemplary natural textures to demonstrate the technique's generality.

This paper introduces an edge detection process based on area morphology. Area open-close and area close-open operators are used to generate scaled image representations for feature extraction. The edges are defined by the boundaries of the scaled objects in the area-filtered images. From the area open-close and close-open operators, thin closed contours suitable for image segmentation are produced. The edge maps allow exact specification of the minimum area for the extracted regions and are Euclidean invariant and causal through scale space. Results are given that demonstrate the effectiveness of the area operator-based edge detection. In contrast to traditional edge detectors, edge detection via area morphology provides well-localized boundaries and does not require thresholding.

We explore the application of area morphology to image classification. From the input image, a scale space is created by successive application of an area morphology operator. The pixels within the scale space corresponding to the same image location form a scale space vector. A scale space vector therefore contains the intensity of a particular pixel for a given set of scales, determined in this approach by image granulometry. Using the standard k-means algorithm or the fuzzy c-means algorithm, the image pixels can be classified by clustering the associated scale space vectors. The scale space classifier presented here is rooted in the novel area open-close and area close-open scale spaces. Unlike other scale generating filters, the area operators affect the image by removing connected components within the image level sets that do not satisfy the minimum area criterion. To show that the area open-close and area close-open scale spaces provide an effective multiscale structure for image classification, we demonstrate the fidelity, causality, and edge localization properties for the scale spaces. The analysis also reveals that the area open-close and area close-open scale spaces improve classification by clustering members of similar objects more effectively than the fixed scale classifier. Experimental results are provided that demonstrate the reduction in intra-region classification error and in overall classification error given by the scale space classifier for classification applications where object scale is important. In both visual and objective comparisons, the scale space approach outperforms the traditional fixed scale clustering algorithms and the parametric Bayesian classifier for classification tasks that depend on object scale.

The paper details a multiscale clustering technique for document page segmentation. In contrast to existing hierarchical (coarse-to-fine), multi-resolution methods, this image segmentation technique simultaneously uses information from different scaled representations of the original image. The final clustering of image segments is achieved through a fuzzy c-means based similarity measure between vectors in scale space. The segmentation process reduces the effects of insignificant detail and noise. Furthermore, object integrity is preserved in the segmentation process.

A significant drawback of the symmetry set evaluation for a 2D shape using the wave diffusion process1 is its slow execution time caused, in large part, by the diffusion step. We first recall the need for a diffusion process. A parallel implementation of the wave diffusion algorithm on a transputer network is presented. A faster alternative approach to detect the symmetry set, which we call the normal transform, and which is similar to the wave propagation mechanism, is described.