Object Contour Vectorization Methods

Images are graphical representations of statistical data and they help a lot in presenting data graphically. Images’ are the best option to deal with different types of data easily and effectively. It’s universally proved that videos and images state data effectively and it’s very easy to deliver any message with the aid of images and graphical presentation. Different methods, techniques are involved in dealing with graphics and images.

In this fast-moving world, it has been observed that this world is depending upon graphics, and using graphics one can easily influence other’s minds. Image editing involved different techniques and different operations. Image editing is a process of image altering.

There are different software and different illustrators are available to deal with image editing and altering. These tools and software help a lot in enhancing, creating, and developing new images as well as alteration of images. In computers, an image is always stored in a form of a grid or pixels. Raster images are always stored in form of pixels in computers. Pixels contain color, brightness, and sharpness elements which give a new and unique, bright effect to images. These pixels help a lot in giving new shapes and brightness to pictures in many ways. Different illustrators and a number of algorithms are used to find out different things and different angles of pictures which are widely used for different purposes in image processing.

Different software like adobe illustrates used to produce vector images, I this world, people usually like vector images as they are easy to modify and reproduce, they are easy to maintain and easy to modify, software like adobe illustrators helps a lot and give a new way of producing new and unique images in form of vector images.

As, nowadays, digital cameras are very common and they have become very popular these days, due to their popularity, image-altering software and programs are readily available.

Different software and programs are readily available which help a lot in creating images and their vectorization, some of them are listed below: Ability Photo paint, ACD Canvas (formerly Deneba Canvas), Adobe Fireworks, Adobe Photoshop, Adobe Photoshop Elements, Alias Sketchbook, Aperture, Arc Soft Photo Studio, Art Rage Coded Color, Corel Painter, Corel Painter Essentials, Corel Paint Shop Pro Photo, Corel Photo-Paint cosmigo Pro Motion, Dig image Arts Color It!, ERDAS IMAGINE, Graphic Converter, ReaSoft Image Converter, Helicon Filter Home/Pro, iPhoto, LView Pro, Manga Studio, Microsoft Digital Image (formerly Microsoft Picture It!), Microsoft Office Picture Manager, Microsoft Paint, MST Viewer, Naked light, Open Canvas, Photogenic, Photo Mechanic, Photo Perfect, PicMaster, Pictor Paint, PixBuilder Photo Editor, Pixel image editor (formerly Pixel32), Project Dogwaffle, SAI.

Most image editors work on compression technologies that reduce file size and make them transferable. This compression technique is used to transfer images, it saves time and money.

Different illustrators and different software’s uses following terminologies: Selection, Layers, Image size alteration, Cropping an image, Histogram, Noise removal, Removal of unwanted elements, Selective color change, Image orientation, Perspective correction and distortion, Lens correction, Sharpening and softening images, Selecting and merging of images, Slicing of images, Special effects, Change color depth, Contrast change and brightening, Color adjustments, Printing, in computers, images are usually stored in form of grid and pixels.

These pixels are really important in producing and creating new and unique graphics. Image contour is a process of defining area and axis of an image to display contour plot of the data in an intensity image, this process is use to set up axis of image so their orientation and aspect ratio match.

There are number of methods which are used for object vectorization in images all of them have own specification and according to which results also vary. There are number of methods, some of them are listed below:

• Image vectorization using optimized gradient meshes
• A randomized knot insertion algorithm for outline capture of planar images using cubic spline
• Vector field contours
• Artistic thresholding
• MAC: Magneto static Active Contour Model
• Gradient Vector Flow Fast Geometric Active Contours
• Shape-based calculation and visualization of general cross-sections through biological data
• Gradient Vector Flow Fast Geodesic Active Contours
• Detection of Calculi Using Active Contour Method
• Segmentation of Discrete Vector Fields
• A Charged Geometric Model for Active Contours
• Shadow volume reconstruction from depth maps
• Degraded character image restoration using active contours: a first approach
• Accurate object contour tracking based on boundary edge selection

Image vectorization using optimized gradient meshes

Nowadays, gradient meshes are widely used as a powerful vector graphic representation for drawing multicolor mesh objects with smooth transition. Tools like Abode Illustrator and Corel CorelDraw, usually they allow a user to create gradient meshes manually even for photo-realistic vector arts, which can be easily edited, stylized and animated and modified by following few simple steps. In this approach an easy-to-use interactive tool is used which is commonly known as optimized gradient mesh, to semi-automatically with the aid of which a user quickly create gradient meshes from a raster image.

This approach is use to obtain the optimized gradient mesh by formulating an energy minimization problem. By following this method a user can easily and interactively specify few vector lines for guiding the mesh generation. The output optimized gradient mesh is an editable and scalable mesh which would have taken many hours for a user to manually create(Chenay,2002).

A randomized knot insertion algorithm for outline capture of planar images using cubic spline

In this method, a very effective algorithm is used which produces effective results as an output image. This technique based upon curve fitting using cubic splines. The technique has various phases including extracting and pointing outlines of images, this method detects corner points from the detected outline, for further image processing in addition it deals with extra knot points if needed. The last phase of this technique makes a significant and valuable contribution by making the technique automated. This approach follows the idea of knot insertion in a randomized manner for effective results. The proposed algorithm is an iterative one which repeat itself unlimited times. The algorithm proposed is computationally efficient as compared to least square approach (Wang,2004).

Vector field contours

This method is used to define 3D vector fields and use them as an interactive flow visualization tool for reproducing image. Although contours are always well-defined and properly drawn so they are commonly used for surfaces and 3D scalar fields, they have no straightforward extension in vector fields, due to which it has become easy to define vector fields (Buck and Purcell,2004). This approach is widely used to extract and visualize specific stream lines which show the most similar and relevant behavior to contours on surfaces.

By this way, the vector field contours are a particular set of isolated stream line segments that depend on the view direction and few additional parameters. This method presents an analysis of the usefulness of vector field contours by demonstrating applications to linear vector fields. In order to achieve interactive visualization, this method develops an efficient GPU-based implementation for real-time extraction and rendering of vector field contours. This approach shows the potential of applying it to a number of example data sets from which desirable results are obtained (Ebert and Rheingans, 2000).

Artistic thresholding

This method main focus the problem of depicting continuous-tone images which uses only black and white. Earlier and traditional solutions to this problem include half toning, which focus on defining tones, and drawing lines, with respect to edges (Santella, 2002). In this approach a new technique is introduced known as “artistic thresholding”. Artistic thresholding is a technique that attempts to depict forms in an image. This technique applies segmentation to a source image and constructs a planar subdivision which helps in capturing segment connectivity (Gooch, A. 2004). A strong thresholding algorithm is working behind this technique, it is a combinatorial optimization over this graph. The optimization is controlled by parameters that can be tuned to achieve different artistic and delicate styles.

MAC: Magneto static Active Contour Model

In this method/ model an external force field which mainly focuses on magneto statics and hypothesized magnetic interactions between the two important components i.e. active contour and object boundaries is used. The main role of following method is the interaction of its forces which helps in improving the active contour in capturing complex geometries and dealing with difficult initializations, it eliminates weak edges and broken boundaries.

The proposed method is mainly focuses in achieving high improvements when compared against six well-known and state-of- he-art shape recovery methods, including the geodesic snake, the combined geodesic and GVF snake, and the charged particle model. This method has great results and considered one of the best object counter vectorization methods (Majid, 2008).

Gradient Vector Flow Fast Geometric Active Contours

This method proposes an edge-driven bidirectional geometric flow for boundary extraction. This approach combines the geodesic active contour flow and the gradient vector flow external force for image. The resulting motion equation produces a level set formulation which can deal with topological changes and important shape deformations. Efficient numerical facts are used in the implementation to create flow which exhibits robust behavior and produces fast convergence rate (Graham, 1995).

Shape-based calculation and visualization of general cross-sections through biological data

This method presents a solution image interpolation problem for producing, high precision images, and an arbitrary cross-section from different levels of slices. Shape-based interpolation is specially design to allow the interpolation process to be influenced by the shape of the object. It is firstly used by creating reconstructed shapes of the objects from levels of slices. Vectors ensure the exact image interpolation over the outline of objects.

Correspondence vectors are associated to every image point within the new cross-section in process, and deal with image points from start to end of vectors which are linearly interpolated. Points and areas of images where the correspondence vectors cannot be treated separately then there is a need of dealing correspondence points differently. Results are presented based on biological data which are exclusively obtained using an optical microscope. The data and results indicates that this approach is one of the best approach for object vectorization and this approach uses statistical and real facts ,data for producing awesome results(H. Yamashita,1997).

Gradient Vector Flow Fast Geodesic Active Contours

In this method a new front propagation flow for boundary extraction is proposed. The framework this technique follows inspired by the active contour model which leads a paradigm which deals initial curve position effectively. This approach makes vectorization easy, simple and attractive, this approach makes use of a recently introduced external boundary force, the gradient vector which refers to a spatial diffusion of the boundary information. According to the proposed approach, the traditional boundary attraction term is replaced in this new proposed approach. With a new force which guides the propagation to the object outlines from both sides.

This flow is deployed using a unique approach, a level approach which allows dealing naturally with topological and graphical changes and is important for shape deformations. In addition, the level set motion equations are usually deployed using a newly introduced numerical based facts and scheme, the Additive Operator Splitting Schema (AOS) which has recorded as a fast convergence rate and stable behavior. This approach produces best results while using real images, results provides fast convergence rate which is known as the key point of this method (Ramesh, 2001).

Detection of Calculi Using Active Contour Method

It is very well known that active contours are widely used for image vectorization. Active Contours are widely used in computer for image processing and image processing applications, they are widely used t locate object boundaries and in image recognition process. In this method Active contour and Gradient Vector models are implemented and study produces co-operative results with respect to image resolutions and image boundaries and robustness to changes of model parameters and user defined parameter values (Karibasappa, 2007).

Segmentation of Discrete Vector Fields

This approach proposes an approach for 2D vector field segments which highly based on the Green function and different normalized cut. This approach is highly inspired by discrete Hodge Decomposition like discrete vector field which can be broken down into different number of different size components. Basically it can be broken down into three components, three simpler components, namely curl-free, divergence-free, and harmonic components.

This approach uses Green Function Method (GFM) which can also be used to approximate the curl-free and the divergence-free components to achieve vectorization goal for the vector field segmentation. The final and complete segmentation curves represent the boundaries and outlines of the influence region of components which are obtained from the optimal vector field segmentations. These curves are fully composed of piecewise smooth contours or streamlines. This approach is applicable on both linear and nonlinear discrete vector fields. Experiments and results using this approach have proved vital and great results as compare to other approach (Wijk, 2001).

A Charged Geometric Model for Active Contours

This model/approach presents a new and model based on charged particle dynamics and geometric contour propagation. It detects object boundaries with a charged active contour that propagates under the forces of an image-based electric field. This approach makes use of layers level set representation to allow major changes to be handled naturally. Also, this approach develops the centre of divergence concept within electric field.

Shadow volume reconstruction from depth maps

Nowadays, mostly graphics hardware are used to generate shadows in different ways i.e. using the shadow volume or shadow map techniques etc. The shadow volume technique demands access to a complete scène of /as a polygonal model, and it requires proper handling of plane clip correctly and efficiently which is quiet difficult; on the other hand, accurate shadow maps require high-precision texture map data representations with complete specifications, but they are not widely supported.

This method presents a hybrid approach of the shadow map and shadow volume which do not have such difficulties and rendering problems. In this approach, the scene is rendered from the point of light source and a depth is recovered. Edge detection and outlining of an image is a template-based reconstruction technique which are widely used to generate a global shadow volume boundary surface, after creating boundary surface pixels in shadow are always be marked using one-bit stencil buffer and a single-pass rendering of the shadow volume boundaries. This technique give excellent results after image gets reconstructs (BERGERON, P. 1986).

Degraded character image restoration using active contours

This method uses active contours for the reconstruction of different character images. This technique was introduced 15 years back by Kass and have been very popular and still in use since then for segmentation purpose or for objects detection or boundary detections in different kind of natural images and real images, but they have never used on documented images. This approach widely deals with all aspects of image countering and altering with different angles and points (PEAIRS, 1993).

Accurate object contour tracking based on boundary edge selection

This method deals with edge detection, mainly boundary edge detection depends on two steps:

1. it removes background edges with the aid of edge motion, and from the output of the previous step,
2. it selects boundary edges using a normal direction derivative of the tracked contour.

Tracking mainly depends upon on the reduction of irrelevant edges from the image, in order to remove irrelevant edges different techniques and ways are being used. This method is highly cost effective in a list of all other methods (Kass, 2003).

Advantages of Object contour vectorization methods

Object Counter Vectorization methods helps a lot in dealing with both documented as well as real images, in computers, raster images are stored in form of grid or pixels. These methods helps a lot in creating boundary lines and detection points plus it helps a lot in detecting edge points which are always required for further image processing. These points are also useful for video processing tracking the complete counter of objects. In some methods, counter tracking algorithm which gives best and accurate results. These methods help a lot in detecting points of whole images and they give best result in real images. Some method gives perfect results on document images.

Methods of probability models

This graphic field is now enhanced a lot and number of things is involved in it nowadays. Following are some methods widely known as probability models:

• A Hybrid System Using Multiple Cyclic Decomposition Methods and Neural Network Techniques
• Modeling Propagation Dynamics
• Mathematical Modeling for the Design of an Edge Router
• Quantitative Verification of Projected Views Using a Power Law Model of Feature Detection
• A Prediction Model of DoS Attack’s Distribution Discrete Probability
• Multilayered 3D LiDAR Image Construction Using Spatial Models in a Bayesian Framework
• Robust Image Segmentation Using Resampling and Shape Constraints
• Change Detection in a 3-d World
• CRF-driven Implicit Deformable Model
• BME : Discriminative Density Propagation for Visual Tracking
• Spatial Random Tree Grammars for Modeling Hierarchal Structure in Images with Regions of Arbitrary Shape

A Hybrid System Using Multiple Cyclic Decomposition Methods and Neural Network Techniques

There are a few preprocessing methods (i.e. ARMA outputs as time domain filters, and Fourier transform or wavelet transform as time-frequency domain filters) for handling different series of images. This series describes different types of restoration points in picture which gives complete ease of finding soft and nice points of an image. In This approach different things are engaged in respect of finding boundary edge detections.

Quantitative Verification of Projected Views Using a Power Law Model of Feature Detection

As, it has been observe that in different images conventional approaches to the construction of same models of visual appearance for image features are non-quantitative, so there was a need of some method which surely deals with this issue. Depending upon images outline and edge detection. This approach outlines a quantitative method for verification of 3D objects’ predicted edge features in images, which surely incorporates both the effects of image noise and local image structure. This method has been validated on multiple and upgraded views of man-made objects constructed from a variety of materials (Simen, 2008).

Change Detection in a 3-d World

This method mainly focuses of detecting changes in a 3-D scene from a huge sequence of images taken or catch by cameras with arbitrary point but with a known and exact pose. Assumption of this method follows no prior knowledge of the state of normal appearance and geometry of object surfaces, and it also assumes that abnormal changes can occur in any image of the sequence at any point if the image is taken from camera.

This method is first address the change detection problem in a general and exact framework. There are number of change algorithms are available which deals with detecting changes in a series of images. Existing change detection algorithms which exploit multiple image viewpoints and then detect only major and motion changes an secondly they assume a planar geometry which couldn’t cope effectively with appearance changes due to different 3-d scene geometry issues(ego-motion parallax).

This method or approach presented here is designed to manage the complications of unknown surfaces and sometimes different change surfaces by maintaining a 3-d voxel-based model, where probability distributions of different surfaces occupancy and image appearance are always taken in each voxel. The probability distributions of each voxel are always continuously gets updated when new images arrived.

The key point and main question involved in this is about convergence. this joint estimation problem can only be answered by a formal proof based approach and real world facts on realistic assumptions about the nature of real world scenes and images. A huge number of experiments are available using this approach that evaluates all major and minor changes in image from a series of image it also produces detection accuracy under laboratory-controlled conditions as well as aerial reconnaissance and real world scenarios based on real facts and representation (Mundy, 2008).

CRF-driven Implicit Deformable Model

This approach based on topology independent solution for segmenting objects with texture patterns at any scale, it deals different type of images irrespective of their resolution etc, it deals at any scale, using an implicit deformable model driven by Conditional Random Fields (CRFs). This model integrates region and edge information as image driven terms, which further helps in improving segmentation and vectorization of images at different scales whereas the probabilistic shape and internal (smoothness) terms use representations similar to the level-set based methods.

The evolution of this particular model is solved and known as a MAP estimation problem, in this problem it targets conditional probability, and decomposed it into the internal term and the image-driven term for further and detail processing. In addition this approach uses discriminative CRFs at two different scales, i.e. pixel-and patch-based, these both scales are used to obtain smooth probability fields in an image which highly based on the corresponding image features and edge points.

There are some advantages and benefits of this particular approach, some of the are listed below:

1. it integrates CRFs with implicit deformable models in a tightly coupled scheme, in order to achieve perfect and smooth results
2. in this approach use of CRFs avoids ambiguities in the probability fields, for perfect segmentations(Gemen,1984).
3. Handling plays an important role and particularly this approach deals proper handling of local feature variations by updating the model inner statistics and process them at different scales in order to achieve best, smooth and exact results.
4. As, this approach is a topology based approach and focuses on independence from the topology. This approach is applied on number of images at different scales and its performance is much better than any other probability methods. This approach is applied on wide range of images, including zebra and cheetah examples plus left and right ventricles in cardiac images and on all images this approach has given marvelous results.

BME: Discriminative Density Propagation for Visual Tracking

This method is new approach which introduces BME, a Conditional Bayesian Mixture of Experts Markov Model, for obtaining best of the best results from consistent probabilistic estimates in discriminative visual tracking. This model applies to a number of problems for obtaining its performance and impact rating.

This model mainly deals problems of temporal and uncertain inference and this model also represents bottom-up counterpart of pervasive generative models estimated based on filtering or particle filtering which plays an important part for obtaining perfect and absolute results. Rather than inverting a non-linear generative observation model, in this model a user will learn to combinable predict complex state distributions directly to encode image observations and image calculations typically this model works on a bag-of-feature global image histograms on both computed over regular spatial grids and sometimes irregular spatial grids.

These grids are always integrated in a conditional graphical model in order to achieve smoothness in results and allow a principled uncertainty management. Existing algorithms have showed great results but there was a lack of smoothing effect in images due to which there was a need of some model which mainly focus this problem, BME model solved this issue up to high extent. Different algorithms usually combine sparsity, mixture modeling, and non-linear dimensionality reduction for efficient and perfect computation in high-dimensional plus multi dimensional continuous state spaces. There are number of advantages of this approach it produces smooth and effective results. Mainly this method deals in three areas

1. It establishes the density propagation rules for discriminative and absolute inference in continuous as well as temporal chain models for effective and desirable results.
2. This method propose flexible supervised and unsupervised algorithms for learning feed forward, multi valued contextual mappings (multimodal state distributions) which highly based on compact as well as conditional Bayesian mixture of experts and already proven models.
3. This model validates the framework and area of image for reconstruction of 3d human motion in video scenes. This model has tested on both real and motion capture-based scenes which shows significant and effective performance gain by competing nearest-neighbor, regression, and structured prediction methods. This approach has given best results and proved as one of the best and effective technique which gives dynamic aspect of the contour.

All above methods deals probability and give dynamic aspects t counter but some limitations are involved in methods. Any method or approach if applied properly it surely give excellent results. For effective and best results it is necessary to apply themes, methods, approaches, models keeping an eye on all aspects of object and desirable results. Any mishandling and misunderstanding effects lead your research and observations towards downfall.

There are number of algorithms/methods which deals with image vectorization effectively and if properly implemented it gives best and effective results. Most of the algorithm uses different approaches and ways, there is always a need of understanding their objectives clearly and then implement it properly following complete sets of rules and limitations. Assumptions also help in driving exact and accurate results.

References list

Cheney, A. S. 2002. Object-based image editing. In Proceedings of SIGGRAPH, 777–784.

Wang, W.2004, Fitting B-Spline Curves to Point Clouds by Squared Distance Minimization, HKU CS Tech Report TR-2004-11.

I. Buck and T. Purcell, 2004, A toolkit for computation on gpus. In GPU Gems, pages 621–363.

D. Ebert and P. Rheingans, 2000, Volume Illustration: Non-Photorealistic Rendering of Volume Models. In IEEE Visualization, pages 195–202.

Santella, A. 2002. Stylization and abstraction of photographs. ACM Trans. Graph. 21, 3, 769–776.

Gooch, B., Reinhard, E., and Gooch, A. 2004. Human facial illustrations: Creation and psychophysical evaluation. ACM Trans. Graph. 23, 1, 27–44.

Majid Mirmehdi, 2008. “MAC: Magneto static Active Contour Model,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 30, no. 4, pp. 632-646.

J. Graham, 1995, Active Shape Models Their Training and Application Computer Vision and Image Understanding, vol. 61, pp. 38-59.

H. Yamashita, 1997, “Shape-based calculation and visualization of general cross-sections through biological data,” iv,pp.2, First International Conference on Information Visualization (IV’97).

Ramesh, 2001, “Gradient Vector Flow Fast Geodesic Active Contours,” iccv, pp.67, Eighth International Conference on Computer Vision (ICCV’01) – Volume 1.

K. Karibasappa, 2007, “Detection of Calculi Using Active Contour Method,” iccima,pp.426-430, (ICCIMA 2007).

J.J.V. Wijk, 2001,“A Phase Field Model for Continuous Clustering on Vector Fields,” IEEE Trans. Visualization and Computer Graphics, vol. 7, pp. 230-241.

BERGERON, P. 1986. A general version of Crow’s shadow volumes. IEEE Comput. Graph. Appl. 6, 9, 17-28.

M. PEAIRS, 1993 “Image Continuation” Proc. Of the 2nd ICDAR, Tsukuba (Japan), pp.53–57.

Kass, M.2003, active contour models. Int. J. Computer Vision. v1 i4. 321-331.

Simon Coupe, 2008 “Quantitative Verification of Projected Views Using a Power Law Model of Feature Detection,” crv,pp.352-358, 2008 Canadian Conference on Computer and Robot Vision.

Joseph L. Mundy, 2008 “Change Detection in a 3-d World,” cvpr,pp. 1-6, IEEE Conference on Computer Vision and Pattern Recognition.

Gemen, 1984, Stochastic Relaxation, Gibbs Distributions, and Bayesian Restoration of Images IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 6, pp. 721-741.

Z.Y. Zhang, 1995, Parameter Estimation Techniques: A Tutorial with Application to Conic Fitting technical report, INRIA, Sofia, France.