Binnur Kurt

License Plate Character Segmentation Based on the Gabor Transform and Vector Quantization

This paper presents a novel algorithm for license plate detection and license plate character segmentation problems by using the Gabor transform in detection and local vector quantization in segmentation. As of our knowledge this is the first application of Gabor filters to license plate segmentation problem. Even though much of the research efforts are devoted to the edge or global thresholding-based approaches, it is more practical and efficient to analyze the image in certain directions and scales utilizing the Gabor transform instead of error-prone edge detection or thresholding. Gabor filter response only gives a rough estimate of the plate boundary. Then binary split tree is used for vector quantization in order to extract the exact boundary and segment the plate region into disjoint characters which become ready for the optical character recognition.

Robust Face Alignment for Illumination and Pose Invariant Face Recognition

In building a face recognition system for real-life scenarios, one usually faces the problem that is the selection of a feature-space and preprocessing methods such as alignment under varying illumination conditions and poses. In this study, we developed a robust face alignment approach based on Active Appearance Model (AAM) by inserting an illumination normalization module into the standard AAM searching procedure and inserting different poses of the same identity into the training set. The modified AAM search can now handle both illumination and pose variations in the same epoch, hence it provides better convergence in both point-to-point and point-to-curve senses. We also investigate how face recognition performance is affected by the selection of feature space as well as the proposed alignment method. The experimental results show that the combined pose alignment and illumination normalization methods increase the recognition rates considerably for all feature-spaces.

Image Compression Based on Centipede Model

We present an efficient contour based image coding scheme based on Centipede Model. Unlike previous contour based models which presents discontinuities with various scales as a step edge of constant scale, the centipede model allows us to utilize the actual scales of discontinuities as well as location and contrast across them. The use of the actual scale of edges together with other properties enables us to reconstruct a better replica of the original image as compared to the algorithm lacking this feature. In this model, there is a centipede for each edge segment which lies along the segment and the gray level variation across an edge point is represented by the difference between footholds and distance between left and right feet of the centipede. We obtain edges by using the recently introduced Generalized Edge Detector (GED) [1] which controls the scale and shape of the filter, providing edges suitable to the application in hand. The detected edge segments are ranked based on the weighted sum of the length of the segment, mean contrast and standard deviation of gray values on the segment. In our scheme, the compression ratio is controlled by retaining the most significant segments and by adjusting the distance between the successive foot pairs. The original image is reconstructed from this sparse information by minimizing a hybrid energy functional which spans a space called Λτ-space. Since the GED filters are derived from this energy functional, we utilized the same process for detecting the edges and reconstructing the surface from them. The proposed model and the algorithm have been tested on both real and synthetic images. Compression ratio reaches to 180:1 for synthetic images while it ranges from 25:1 to 100:1 for real images. We have experimentally shown that the proposed model preserves perceptually important features even at the high compression ratios.

Gradient-based shape descriptors

This paper presents two shape descriptors which could be applied to both binary and grayscale images. The proposed algorithm utilizes gradient based features which are extracted along the object boundaries. We use two-dimensional steerable G-Filters (IEEE Trans Pattern Anal Mach Intell 19(6):545–563, 1997) to obtain gradient information at different orientations and scales, and then aggregate the gradients into a shape signature. The signature derived from the rotated object is circularly shifted version of the signature derived from the original object. This property is called the circular-shifting rule (Affine-invariant gradient based shape descriptor. Lecture notes in computer science. International workshop on multimedia contents Representation, Classification and Security, pp 514–521, 2006). The shape descriptor is defined as the Fourier transform of the signature. We also provide a distance measure for the proposed descriptor by taking the circular-shifting rule into account. The performance of the proposed descriptor is evaluated over two databases; one containing digits taken from vehicle license plates and the other containing MPEG-7 Core Experiment and Kimia shape data set. The experiments show that the devised method outperforms other well-known Fourier-based shape descriptors such as centroid distance and boundary curvature.

Affine invariant gradient based shape descriptor

This paper presents an affine invariant shape descriptor which could be applied to both binary and gray-level images. The proposed algorithm uses gradient based features which are extracted along the object boundaries. We use two-dimensional steerable G-Filters [1] to obtain gradient information at different orientations. We aggregate the gradients into a shape signature. The signatures derived from rotated objects are shifted versions of the signatures derived from the original object. The shape descriptor is defined as the Fourier transform of the signature. We also provide a distance definition for the proposed descriptor taking shifted property of the signature into account. The performance of the proposed descriptor is evaluated over a database containing license plate characters. The experiments show that the devised method outperforms other well-known Fourier-based shape descriptors such as centroid distance and boundary curvature.

Active appearance model based face recognition

In this study, we present an approach for illumination invariant face recognition based on active appearance model (AAM) and the Gabor transform. Salient face components (i.e., eyes, nose, and chin) are extracted automatically by using AAM. We have showed that edges originating from object boundaries are far less susceptible to light source changes. Here, we propose a contour detector which only collects contours originating from object boundaries and eliminates the others arising from texture. We have used hill representation[14] of images which are obtained by spanning a membrane surface through the detected edges. The hill representation increases AAM localization accuracy compared to the case where RGB values are used alone. We demonstrate how successful the proposed algorithm is on real face images obtained under various lighting conditions.

Two dimensional generalized edge detector

Detecting edges in images is one of the most challenging issues in computer vision and image processing due to lack of a robust detector. Gokmen and Jain (1997) have obtained an edge detector called the generalized edge detector (GED), capable of producing most of the existing edge detectors. The original problem was formulated on a two-dimensional hybrid model comprised of the linear combination of membrane and thin-plate functionals. The smoothing problem was then reduced to the solution of two-dimensional partial differential equations (PDE). The filters were obtained for the one-dimensional case assuming a separable solution. This study extends edge detection of images in /spl lambda//spl tau/-space to two-dimensional space. The two-dimensional extension of the representation is important since the properties of images in the space are best modeled by two-dimensional smoothing and edge detector filters. Also since GED filters encompass most of the well-known edge detectors, two-dimensional versions of these filters could be obtained. The derived filters are more robust to noise when compared to the previous one-dimensional filtering scheme in the sense of FOM (figure of merit), missing and false alarm characteristics. Experimental results on synthetic and natural images are presented, including an analysis of the introduced two-dimensional edge detector filters and the behaviour of the detected edges through the /spl lambda//spl tau/-space.

Goal oriented edge detection

In many vision applications, there is a great demand for an edge which can produce edge maps with very different characteristics in nature, so that one of these edge maps may meet the requirements of the problem under consideration. Unfortunately it is not evident how to choose the desired or the optimum edge maps from these solutions that the edge detector offers. The proposed solutions are usually too general that cannot be easily adapted to the application needs by tuning edge detection parameters. One edge detector that we have studied in this study is Generalized Edge Detector which is capable of producing edges with very different characteristics. Although the edge maps based on this representation are reasonable, no one set of scale parameters alone yields a solution close to the desired edges. In this study, we have developed powerful edge operates and have used them under a goal-based edge detection framework. Proposed framework is a two-stage process. First, user marks some pixels in the database as edge and non-edge pixels. Then feature vectors comprised of filter responses to G-Filters at different scales are extracted at these marked pixels. Edge detection problem is imposed as two-class classification problem. Classifier itself is not adequate to extract desired edges for the application under consideration. In the second stage continuous edges are treated as one contour. Then contours are matched with the contours in the training set. Only matched contours are kept and the other contours are eliminated. The purpose of the first stage is to keep only prominent edges and remove irrelevant edges with respect to the application. The classifier decides which discontinuity is prominent or irrelevant. Experimental studies on real license plate images show that the proposed edge detector can successfully detects edges only on license plate regions.

Affine Invariant Shape Descriptors

This paper presents affine-invariant shape descriptor which could be applied to both binary and gray-level images. The proposed algorithm utilizes gradient based features which are extracted along the object boundaries. We use two-dimensional steerable G-Filters ([1]) to obtain gradient information at different orientations and scales. We aggregate the gradients into a shape signature. The signature derived from the rotated object is circularly shifted version of the signature derived from the original object. This property is called the circular-shifting rule ([2]). The shape descriptor is defined as the Fourier transform of the signature. We also provide a distance definition for the proposed descriptor taking the circular-shifting rule into account. The performance of the proposed descriptor is evaluated over the databases containing digits taken from vehicle license plates. The experiments show that the devised method outperforms other well-known Fourier-based shape descriptors such as centroid distance and boundary curvature.

View independent robust license plate recognition system

In this study, Gabor kernels, which are tuned for license plate texture, are used for detection of license plate position. In order to expand the application of license plate recognition into various fields, it is necessary to develop an algorithm qualified to handle more deformable plates. Finally we use affine rectification to recover any deformation on the plate region of rectangular shape caused by an improper camera viewing parameters. Experimental results for the license plate image database including plates from various countries (i.e., fonts and shapes), aspect ratios, and sizes demonstrate the great performance of the proposed method.