Yujuan Sun

Fast 3D face reconstruction based on uncalibrated photometric stereo

This paper proposes a fast algorithm for three-dimensional face reconstruction using uncalibrated Photometric Stereo. With a reference face model, lighting parameters are estimated from input face images lighted by unknown illumination, which can be used in classical photometric stereo to estimate surface normal and albedo. The estimated results are used in turn to refine the lighting parameters until an optimal estimation of the surface normal is achieved. Differing from traditional optimization algorithms, the iteration method used in this paper is a unified process thus results accurate lighting estimation. The proposed method relaxes lighting constraints and simplifies the image acquisition procedure. The reconstructed results tested on YaleB and BU3D databases show the effectiveness of our method.

Saliency detection using quaternionic distance based weber local descriptor and level priors

In this paper, a novel and efficient framework by exploiting Quaternionic Distance Based Weber Local Descriptor (QDWLD) and object cues is proposed for image saliency detection. In contrast to the existing saliency detection models, the advantage of the proposed approach is that it can combine quaternion number system and object cues simultaneously, which is independent of image contents and scenes. Firstly, QDWLD, which was initially designed for detecting outliers in color images, is used to represent the directional cues in an image. Meanwhile, two low-level priors, namely the Convex-Hull-Based center and color contrast cue of the image, are utilized and fused as an object-level cue. Finally, by combining QDWLD with object cues, a reliable saliency map of the image can be computed and estimated. Experimental results, based on two widely used and openly available database, show that the proposed method is able to produce reliable and promising salient maps/estimations, compared to other state-of-the-art saliency-detection models.

Improved fuzzy clustering algorithm with non-local information for image segmentation

Fuzzy C-means(FCM) has been adopted to perform image segmentation due to its simplicity and efficiency. Nevertheless it is sensitive to noise and other image artifacts because of not considering spatial information. Up to now, a series of improved FCM algorithms have been proposed, including fuzzy local information C-means clustering algorithm(FLICM). In FLICM, one fuzzy factor is introduced as a fuzzy local similarity measure, which can control the trade-off between noise and details. However, the fuzzy factor in FLICM cannot estimate the damping extent of neighboring pixels accurately, which will result in poor performance in images of high-level noise. Aiming at solving this problem, this paper proposes an improved fuzzy clustering algorithm, which introduces pixel relevance into the fuzzy factor and could estimate the damping extent accurately. As a result, non-local context information can be utilized in the improved algorithm, which can improve the performance in restraining image artifacts. Experimental results on synthetic, medical and natural images show that the proposed algorithm performs better than current improved algorithms.

Saliency detection using quatemionic distance based weber descriptor and object cues

In this paper, a simple and efficient method, based on Quaternionic Distance Based Weber Descriptor (QDWD) and object cues, is proposed for saliency detection. Firstly, QDWD, which was initially designed for detecting outliers in color images, is used to represent the directional cues in an image. Meanwhile, two low-level priors, namely the color contrast and center cue of the image, are utilized and fused as an object-level cue. Finally, by combining QDWD with object cues, a reliable saliency map of the image can be computed. Experimental results, based on a widely used and openly available database, show that the proposed method is able to produce promising results, compared to other state-of-the-art saliency-detection models.

An improved genetic algorithm for three-dimensional reconstruction from a single uniform texture image

Three-dimensional reconstruction from a single input image is a very difficult issue, especially for the texture images. Moreover, the unknown lighting parameters also make this problem more complex. In this paper, an improved genetic algorithm has been proposed to reconstruct the 3D shape from a single texture image with similar appearances. The proposed scheme contains three main steps: first, the lighting parameters has been estimated by detecting and analyzing the intensity information of the input texture image; then, the initial surface normal, which can be used as the initial population of generic algorithm, has been calculated by combining the patch matching and stitching method; finally, the improved genetic algorithm incorporating spatial information is implemented, which can search the minimum starting from the surface normals of the neighborhood. Experiment results verified the effectiveness of the proposed method according to realistic visual-perception.

Person Re-Identification with Deep Features and Transfer Learning

Person re-identification is an important technique towards automatic search of a persons presence in a surveillance video. Two fundamental problems are critical for person re-identification:feature representation and metric learning. At present, there are many methods in the study of person re-identification, which has achieved remarkable results. Due to the difference of the data distribution in different scenarios, the performance of the person re-identification in the new scene is significantly decreased. In order to avoid the tedious manual annotation, and to make full use of the original detector and labeled samples, the research of person re-identification based on transfer learning has received more and more attention. Existing approaches adopt a fixed metric for matching all the subjects. In this work, we propose a Feature Net (FN) architecture with Convolution Neural Networks (CNNs) to learn the pedestrian feature, reserved more useful information. And use Cosine distance to measure the each image pairs similarity directly which is more efficient but uncomplicated than others. Our method can be applied to different scenarios and improved the recognition performance. Experiments on the challenging datasets show the effectiveness of our methods, especially on cuhk03 dataset, we achieve the state-of-the-art result.

Reconstruction of normal and albedo of convex Lambertian objects by solving ambiguity matrices using SVD and optimization method

Photometric stereo (PMS) can reconstruct shape and albedo of an object by using multiple images captured under varied illumination directions. However, PMS may fail if light intensity is varied across different images captured under different unknown lighting directions. This paper presents a method that can estimate shapes and albedo of inhomogeneous Lambertian objects with much less constrained lighting conditions, i.e. the illumination directions are unknown and there can be arbitrary combination of different light sources and ambient light; meanwhile the light intensity can be different in different images. By placing a reference object alongside an object, the ambiguous matrix produced by SVD can be estimated effectively. This matrix is then used to generate more accurate shape and albedo. The reconstructed results are further refined using an optimization algorithm. Both synthetic and real objects are used in our experiments and the results show the effectiveness of our method.

Specular reflection removal of ocean surface remote sensing images from UAVs

Images captured by UAVs above the sea surface often contain lots of highlight regions due to the specular reflection of solar radiation on the non-flat sea surface. The existence of a great deal of specular highlight components may cover the objects under the water which is negative to those applications based on the UAV remote sensing images. In this paper, we present a method to remove the specular reflection on the RGB images of ocean surface. The intensity of specular highlight components is much larger than that of diffuse components in the images, simply subtracting the highlight component form the original image will leave a lot of holes. So our method contains two main steps: highlight regions detection and restoration of those regions. We use the method based on the intensity ratio to extract the regions affected by the specular reflection. Then we use the local information around those highlight regions to restore the intensity of those pixels. The experimental results indicate that the proposed method can effectively remove the specular reflection and keep details of ocean surface images.

Three Dimensional Surface Reconstruction Based on the Reference Object and PMS

Three-dimensional surface reconstruction based on photometric stereo requires accurate positioning of the light source or estimation of the lighting parameters, which increases the complexity of the operation in experiments. Moreover, the actual light environment is composed of many kinds of complex optical components, and the accurate lighting parameters cannot be measured. In this paper, a reference object (Lambertian zirconia ball) is used to estimate the lighting parameters of the actual light environment, and then the lighting matrix of the target object can be deduced based on the same light scene of the reference object and the target object. With the estimated lighting matrix, the surface normal of the target object can be quickly computed. The experiment results of the synthetic and real object have verified the effectiveness of the proposed method.

The Research of Texture Transmission Algorithm Combined with Texture Synthesis

Texture synthesis is the hot research topic in the field of computer vision, computer graphics and image processing. Sample-based texture synthesis method has been proposed and becomes a new texture tiling technique with the development of the texture mapping and procedural texture synthesis. Image Quilting stitching algorithm is an ideal algorithm for the texture stitching. In this article, by using the image Quilting algorithm to finish the texture transfer and stitch the objective texture image, whose texture style does not exist in the initial image. In experiments, by transferring the style of the human face image and character image respectively, verified the effectiveness of this proposed algorithm.