In this paper, we first present a non uniform optimal sampling strategy based on mathematical morphology, for the description of image sequences. This sampling scheme is determinist and iterative. At each iteration, a sample position in the sequence, is determined in order to minimize the overall absolute reconstruction error. This procedure is repeated until reaching either an a priori fixed number N of samples, or satisfying a maximal admissible error criterion. The associated sequence reconstruction procedure is described as well. The reconstructed sequence is an approximation of the opened sequence by the family of structuring elements used. The second part is devoted to the elaboration of a complete sequence coding scheme based on this sampling strategy for image description. Finally, simulation results on video sequences illustrate the performances of this new coding scheme.
Segmentation is the process of dividing an image into segments that have similar attributes. There are a variety of traditional segmentation technique in the literature, mainly for 2D image processing applications. 3D segmentation is a much more complex problem. This is even harder for real 3D data sets where the images are degraded by blur and noise, the background is nonuniform and objects do not possess clear cut boundaries. These traditional techniques are usually unsuitable for segmentation of real 3D images. In this paper we present a novel and effective data driven segmentation framework based on a combination of nonlinear restoration and watershedding. The framework is presented and discussed as are experimental results showing its effectiveness in accurately segmenting real 3D images.
The work presented in this article describes a tool for object tracking, notes insertion, and information retrieval, applicable to MPEG-2 sequences. Maximum compliance with the MPEG standard is sought, so the added information is transmitted as side information without affecting the actual video-audio stream as defined in the MPEG-2 standard. Additional processing is added to a standard sequence, allowing for automatic tracking of one object across different groups of pictures. Results show that the proposed algorithm is capable of tracking objects with a good degree of precision. Features are included to alert the human operator when objects disappear, or must be considered lost, due to an excessive change in their shape
This paper describes some 3-D image analysis methods for detecting changes in facial shape arising from surgical treatment. Metric reconstruction of 3-D facial depth maps has been achieved using shape from stereo. The performance of ICP matching for registering separate facial scans of an individual is evaluated using different surface patch geometries and under various noise levels. The technique is then applied to depth map subtraction and to calculation of the symmetry plane. Finally, thin plate spline warping is used to visualize shape changes in the facial mid-line profile due to re-positioning of the mandible.
An improved least squares stereo matching (LSSM) approach is presented for the accurate recovery of 3-D facial shape. The efficiency of this iterative search technique depends crucially on the provision of good initial disparity estimates between search and target windows. Here, we obtain these using an Active Shape Model (ASM) search to first localise important image features, such as the eyes, nose, mouth and facial border. Disparity values are then estimated from the fitted model points and fed into the matching algorithm. This enhances the correspondence search in facial regions of interest resulting in dense depth maps and better reconstruction accuracy. The technique has been validated by metric reconstruction of 3-D facial models for a human subject.
A new divide-and-conquer technique for disparity estimation is proposed in this paper. This technique performs feature matching recursively, starting with the strongest feature point in the left scanline. Once the first matching pair is established, the ordering constraint in disparity estimation allows the original intra-scanline matching problem to be divided into two smaller subproblems. Each subproblem can then be solved recursively, or via a disparity space technique. An extension to the standard disparity space technique is also proposed to compliment the divide-and-conquer algorithm. Experimental results demonstrate the effectiveness of the proposed approaches.
Traditionally, non-linear diffusion processes and watershed segmentation have been well studied for greyscale image segmentation. In this paper we extend their use to colour images. First, we formulate a general definition for a non-linear diffusion process using the concept of an activity image that can be calculated for several image components. Then, we explain how the cleaned activity image is fed through a watershed algorithm yielding the colour image segmentation. Finally, the qualitative performance is illustrated with results for real colour images.
This paper proposes a new modeling method of texture images based on morphological operations. For the texture analysis One of important texture modeling method is the linear prediction. In the linear prediction method, the frequency feature of texture images is put into autore-grssive (AR) model (i.e., all-pole linear filter) On the other hand,we would like to show new models which represent the structural feature of texture images. It is well known, the patter spectrum calculated by morphological operations are utilized the texture analysis, since morphological operations have the ability to extract structural feature. Thus, we attempt to put into structural feature of texture images put into the structural element of morphological operations. We show the effectiveness of the proposed models through the recognition test.
Most adaptive image and signal processing tasks are performed on specialist digital signal processing chips. These devices are highly optimised for efficient computation of the core multiply and accumulate operations required by current algorithms. Attempts to accelerate image processing using these types of algorithms on FPGAs have resulted in few competitive implementations. FPGAs generally fail to realise efficient arithmetic functions except in the most constrained cases such as constant coefficient multipliers [1]. The approach adopted in this paper is based on the use of stack filters which avoid these difficulties by employing completely different algorithms that do not rely on any arithmetic functions. This is then extended to show how these fast algorithms can be used to accelerate the Hough transform.
This paper derives a new tree representation of an image and shows how the tree may be derived from graph morphology and connected-set, alternating sequential, filters. The resulting scale tree forms a pyramid of increasing size objects where the nodes correspond to features of a particular scale. The tree structure itself may be made fairly insensitive to geometrical changes in the image. By parsing the tree and using attributes associated with the nodes, image processing operations such as filtering, segmentation and detection can be performed.
In this paper we present a novel representation for arbitrary surfaces that enables local correspondences to be determined. We then describe how these local correspondences can be used to search for the transformation that best aligns all of surface data. If this transformation is found to align a significant proportion of the surface data then the surfaces are said to have a correspondence.
Many old paintings suffer from the effects of certain physicochemical phenomena, that can seriously degrade their overall visual appearance.Digital image processing techniques can be utilized for the purpose of restoring the original appearance of a painting, with minimal physical interaction with the painting surface. In this paper, a number of methods are presented which can yield satisfactory results. Indeed, simulation results indicate that acceptable restoration performance may be attained, despite the small size of painting surface data utilized.
Among other techniques especially methods working fully automatically are of interest for image retrieval from large databases. Colour histograms proved to be successful in automatic image retrieval, however, their drawback is that all structural information is lost. Therefore we extend the colour histogram approach by features that take into account the relations within a local pixel neighbourhood. By integrating nonlinear functions over the group of Euclidean motion we extract features that are invariant with respect to translation and rotation. In contrast to approaches using linear filtering (e.g. wavelets) or corresponding power spectra (to become invariant) these nonlinear invariants have the potential to be unique with respect to the equivalence class of Euclidean motion. So in invariant feature histograms we combine the advantage of an invariant description (e.g. we only need one histogram for a whole class of transformed images in the database) with the properties of histogram approaches, providing the possibility to find images also by partial views or vice versa or to detect objects also under occlusion.
A combined approach for facial feature extraction and determination of gaze direction is proposed that employs some improved variations of the adaptive Hough transform for curve detection, minima analysis of feature candidates, template matching for inner facial feature localization, active contour models for inner face contour detection and projective geometry properties for accurate pose determination. The aim is to provide a sufficient set of features for further use in a face recognition or face tracking system.
Tools for detecting scene-changes in digital video are discussed in this paper. We describe tools useful for raw video-data as well as for MPEG-2 compressed bit-streams. Our approach for finding shot-boundaries in raw-data is based on comparisons of color histograms. We provide a comparison of several histogram-similarity measures that have been proposed by other researchers. We also propose the use of the cosine measure which outperforms the other measures discussed here. An algorithm for detecting shot boundaries in MPEG-2 compressed sequences is also described. It extends the work presented in [1] by processing P-frames and B-frames in addition to I-frames.
This paper presents a log-polar image representation composed of low-level features extracted using a connectionist approach. The low level features (edges, bars, blobs and ends) are based on Marr’s primal sketch hypothesis for the human visual system [3] and are used as the entry point of an iconic vision system [1]. This unusual image representation has been created using a neural network that learns examples of the features in a window of receptive fields of the image representation.
3D object modeling is an important issue in the construction of virtual environment. This paper presents an automatic modeling approach for the construction of curved object models. The model is generated from the contours on orthographic planes of images taken from known viewpoints, and represented with a set of superquadric primitives. The parameters of model are determined automatically by first utilizing a global fitting procedure where the silhouettes of the object are fitted to the parametric model, and then a local fitting strategy where some deformable parameters are adjusted to the object edges to refine the parametric model. Combining these fitting parameters, the geometric shape description of 3D objects can be obtained.
We propose a pattern classiication based approach for simultaneous 3-D object modeling and segmentation in image volumes. The 3-D objects are described as a set of overlapping ellipsoids. The segmentation relies on the geometrical model and graylevel statistics. The extension of the Hough Transform algorithm in the 3-D space by employing the spherical coordinate system is used for ellipsoidal center estimation. The characteristic parameters of the ellipsoids and of the graylevel statistics are embedded in a Radial Basis Function (RBF) network and they are found by means of unsupervised training. We propose a new robust training algorithm for RBF networks based on-Trimmed Mean statistics. The proposed algorithm is applied for tooth pulpal blood vessel segmentation in a stack of microscopy images.
This paper describes an algorithm to segment stereo images. The key idea is based on the hybrid use of various image information resources about the scene contents. In order to get reliable segmentation result, not only the pixel-orientated luminance distribution of the image but also object contours and disparity information in stereo image pairs are analysed. This algorithm is applied to an interactive multimedia system with an autostereoscopic display to produce dynamic perspectives and to simulate the limited depth of focus mechanism of the human eye.
Grain noise is one of the most common distortions in cinematographic film sequences and is caused by the crystal structure of chemical coating of the film material. The colour sensitive crystals can be considered as three separate populations. Thus noise in the three channels is uncorrelated and similarly noise between frames is uncorrelated. Conversely, the signal (ie the projected view volume) is highly correlated between channels and over time. We shall explore methods of using this constraint to reduce noise within an adaptive filter framework using the popular Widrow-Hopf LMS algorithm. As a film sequence typically includes many moving elements such as actors on a moving background, motion estimation techniques will be used to eliminate as much as possible the effect of greylevel variations on the adaptive filter. An optical flow technique is used to extract pixel motions prior to the application of the noise reduction.
