Image Rendering And Object Identification Based On Images

Image-Based Rendering (IBR) technique is an alternative strategy for constructing virtual scene, and it has become an important research area in Computer Graphics in last decade or so. IBR uses images pre-captured as basic representation of original scene, and the desired scene corresponding to virtual viewpoint is rendered by sampling, warping and re-sampling the captured images. Comparing with the way of Geometry-Based, the scene rendered is natural, and the rendering time is independent of the scene complexity. But so far it still has many problems needed to be improved, the using visual feature to match images automatically is the one of them.Image-Based Pattern Recognition technique is widely used in a lots of fields such as scenery images classification, computer aided medicine diagnoses and metallography analyzing, etc. but the application in food science and industry has rarely seen. To availably distinguish the sorts and ingredients in food made from starches in merchandise field is required, and which is important to food safety, food nutrition and development healthily in economy. Traditionally, it mainly relies on sensory perception in Bureau of Entry-Exit Inspection and Quarantine or Quality food Quality and Technical Supervision, and which is of fallibility or trustless. By using computer system to automatically accomplish this task based on the microstructure features of the starches is an effective method.The image features are the external reflection of object intrinsic attributes, which are also the perceptions of vision and can be used as input of classification by computer. According to this fact, this paper studies the image features adapted to the mentioned problems above. Comer point on edge were selected for IBR technique, since they has highly repeatability being detected across image sequences when images have considerably changes in rotation, translation, scale and illumination. The comer detector based on phase congruency model was employed for feature detection, because the comer set detected by this detector is a strict subset of the edge set, and more suitable for human perception. The statistical features derived from gray-level co-occurrence and fractal dimension have been used to classify the starch-noodles, owing to these features reflect the attribute of starch-noodle.Matching feature point between image pair in novel view synthesis is a key step and difficult. If existing false matching point between two images, ghost-like will appear on the synthesized virtual image. A novel method to point feature matching is proposed in this paper. In which the feature points are extracted by comer detector first, then the feature scale corresponding to a feature point is obtained in scale-space by using LOG, thirdly the feature area that is circularity consisting of feature point and the feature scale are described by using color moment, finally the robustness matching point is obtained by applying RANSAC.Panoramic image mosaics is a basic technique in IBR, and its fundamental problem is making panorama. The image planar mosaics is studied, and an image automatic mosaics algorithm based on matched feature point is proposed in this paper. To begin with, through taking merely feature points on image pair obtain the initial projective transformation matrices. Then, an optimization algorithm is used to optimize the initial projective transformation matrices by using area around each feature point, and complete the precise image mosaics. The proposed algorithm can stably yield high quality mosaics, even in the cases of poor or differential lighting conditions, existences of minor rotations and other complicated displacements between images.Most methods of view synthesis use the original images taken by camera moving around the object or scene, calibrated camera and rectified images for produceing perspectively correct views, and not completely handled visibility in synthesized views. Unlike these methods we realize the view synthesis between original images captured by an uncalibrated, forwards translating handy camera. The synthesis scheme we developed can deal with changes in visibility, produce perspectively correct views and handle the distortion around the circumferences in the synthesized views. In our scheme, the area without corresponding points was first determined by using Fourier transformation in order to deal with distortion problem, within the mutual overlapped area the z-buffer values of matched key points were computed so that the visibility in novel views can be handle correctly, lastly the common points in new views are rendered through epipolar geometry constraint. This method does not require any kind of prior knowledge about camera and scene.According to the microstructure features of starch-noodles made from starches of crop, an approach that identifies the longkou starch-noodle by using computer system automatically based on the technique of pattern recognition is presented. The method consists of three step: 1) take the micrograph of starch-noodles with scanning electron microscopy. 2) extract features of fractal geometry and Gray-Level Co-Occurrence from micrograph. 3) distinguish a sort of starch-noodles by using these combined features as input vector of artificial neural networks. The results of experiments show that it is practicable and effective.