Description: 一个基于支持向量机和傅立叶描述子的字符识别算法:傅立叶描述子表征字符的轮廓特征-One based on support vector machine and Fourier descriptors of the character recognition algorithm: Fourier descriptor contour characteristics characterization characters Platform: |
Size: 5120 |
Author:zzflash |
Hits:
Description: 状是含有高层语义信息的视觉特征，在基于内容的图像检索及图像识别中具有重要的应用价值。有很多种描述子可以描述图像的形状特征，傅立叶描述子可以把二维的图像轮廓信息简化成一维问题进行处理，应用非常广泛。然而自然图像的形状特征通常是杂乱的，有噪声的，提出了一种图像预处理方法，得到净化的形状图像，通过实验研究傅立叶描述子算法提取形状特征的效果。-Abstract Shape is a visual feature which contains intrinsic high-level semantics, and has a great application value in CBIR(Content-Based Image Retrieval) and IR(Image Recognition). There are many descriptors for shape feature. Fourier descriptor predigests 2-demensional image information to 1-demensional signal and be used widely. In fact, the shape of natural image is often messy and noisy. So, this paper proposes a preprocessing method which can clean the noisy shape image, and then researches and analyses the shape feature extraction with Fourier descriptor method with an experiment.
Keywords Shape, Fourier Descriptor, Feature Extraction, CBIR(Content-Based Image R
Platform: |
Size: 102400 |
Author:倪晓雷 |
Hits:
Description: 火灾视觉特征的提取是视觉火灾探测中的关键问题. 我们主要研究色彩、纹理以及轮廓脉动
等特征的提取,并提出一种度量轮廓脉动信息的距离模型,该模型在规格化的傅立叶描述子空间能
够准确地度量这种时空闪烁特征. 实验结果表明,该方法具有比较好的鲁棒性,有助于提高视觉火
灾探测的准确率、降低误报漏报率.-Based on investigating color , text ure and temporal feat ures for vision based fire detection , a
distance model of contour fluct uation between two successive f rames in t he normalized Fourier descriptor s
domain was presented to measure t his time varying contour fluct uation feat ure of flame. The model of
contour fluct uation is effective and robust for fire recognition. To f urt her reduce fal se alarms , several
features ext racted according to color , text ure and the distance model were toget her regarded as a joint
feature vector for artificial neural network to detect fire. Experiment s show t hat the algorithm is effective
and robust , and t hat it is significant for improving accuracy and reducing fal se alarms. Platform: |
Size: 819200 |
Author:陈卿 |
Hits:
Description: 傅立叶描述子是分析和识别物体形状的重要方法之一．利用基于曲线多边形近似的连续傅立叶变换方法 计算傅立叶描述子，并通过形状的主方向消除边界起始点相位影响的方法，定义了新的具有旋转、平移和尺度不变 性的归一化傅立叶描述子．与使用离散傅立叶变换和模归一化的传统傅立叶描述子相比，新的归一化傅立叶描述 子同时保留了模与相位特性，因此能够更好地识别物体的形状．实验表明这种新的归一化傅立叶描述子比传统的 傅立叶描述子能够更加高效、准确地识别物体的形状．-Abstract Shape is a visual feature which contains intrinsic high-level semantics, and has a great application value in CBIR(Content-Based Image Retrieval) and IR(Image Recognition). There are many descriptors for shape feature. Fourier descriptor predigests 2-demensional image information to 1-demensional signal and be used widely. In fact, the shape of natural image is often messy and noisy. So, this paper proposes a preprocessing method which can clean the noisy shape image, and then researches and analyses the shape feature extraction with Fourier descriptor method with an experiment. Platform: |
Size: 349184 |
Author:劳世华 |
Hits:
Description: 基于叶片数字图像的植物识别是自动植物分类研究的热点。但是随着植物种类的增加，传统的分类方法由 于提取的特征比较单一或者分类器结构过于简单，导致叶片识别率较低。为此，本文提出使用纹理特征结合形状 特征进行识别，并且使用深度信念网络构架作为分类器。纹理特征通过局部二值模式、Gabor 滤波和灰度共生矩阵 方法得到。而形状特征向量由 Hu 氏不变量和傅里叶描述子组成。为了避免过拟合现象，使用“dropout”方法训练 深度信念网络。这种基于多特征融合的深度信念网络的植物识别方法-Plant based on digital image recognition is a hotspot of research on automatic classification.But with the increase of plant species, the traditional classification method by the extraction of characteristics or more single classifier structure is too simple, leading to a lower leaf recognition rate.To this end, this paper proposes using the texture characteristics in combination with characteristics of shape, which can identify the belief network architecture and using the depth as a classifier.Texture characteristics by local binary pattern, Gabor filter and gray level co-occurrence matrix method.And shape characteristic vector by Hu s invariant and the Fourier descriptor.In order to avoid over fitting phenomenon, dropout method is used to train deep belief networks.This belief network based on feature fusion depth plant identification method Platform: |
Size: 377856 |
Author:hahah |
Hits:
Search - recognition fourier descriptor