Color Feature Extraction

Source: http://www.xuebuyuan.com/2019161.html

Color feature is the most widely used visual feature in image retrieval, mainly because the color is very relevant to the object or scene contained in the image. In addition, compared with other visual features, the color features have less dependence on the size, orientation and visual angle of the image itself, and thus have high robustness.

Characteristics of color characteristics: color characteristics is a global feature, describes the image or image area corresponding to the surface properties of the scene. The general color feature is a pixel-based feature, where all pixels belonging to an image or image region have their own contribution. Because the color is not sensitive to the direction and size of the image or image area, the color feature does not capture the local characteristics of the object in the image well. In addition, when using only color feature queries, many unwanted images are often retrieved if the database is large. Color histogram is the most commonly used method of expressing color characteristics, its advantage is not affected by the change of image rotation and peace, the further use of normalization can not be affected by the image scale changes, the base disadvantage is not to express the color space distribution information.

Feature extraction and matching method used in color feature

(1) Color histogram

Color histogram is a color feature widely used in many image retrieval systems. It describes the proportions of different colors in the entire image, and does not care about the spatial location of each color, which means that objects or objects in the image cannot be described. Color histograms are especially useful for describing images that are difficult to auto-segment.

Of course, the color histogram can be based on different color spaces and coordinate systems. The most common color space is the RGB color space, because most digital images are expressed in this color space. However, the RGB spatial structure does not accord with people's subjective judgement of color similarity. As a result, color histograms based on HSV space, Luv space, and lab space have been proposed because they are closer to people's subjective understanding of color. Where HSV space is the most commonly used color space for histograms. Its three components represent color (Hue), saturation (saturation), and values (value), respectively.

Calculating the color histogram requires dividing the color space into a number of small color bands, each of which becomes a bin of the histogram. This process is called color quantization (quantization). Then, the color histogram can be obtained by calculating the number of pixels that fall within each small interval. Color quantization has many methods, such as vector quantization, clustering method or neural network method. The most common practice is to divide the various components (dimensions) of the color space evenly. In contrast, clustering algorithms take into account the distribution of image color features throughout the space, thus avoiding the very sparse number of pixels in some bins, making quantization more efficient. Also, if the image is in RGB format and the histogram is in HSV space, we can pre-establish a lookup table between the quantization RGB space and the quantization of the HSV space (look-up
table), which accelerates the calculation of the histogram.

The color quantization method above will produce some problems. Imagine that the color histogram of two images is almost the same, just staggered one bin, if we use L1 distance or Euler distance to calculate the similarity between the two, we will get a very small similarity value. To overcome this flaw, it is necessary to consider the similarity between similar but different colors. One method is to use a two-time distance. Another way is to filter the color histogram in advance, that is, the pixels in each bin contribute to several adjacent bins. The similarity between similar but different colors also contributes to the similarity of the histogram.

Choosing the right Color cell (that is, the bin of the histogram) and the color quantization method are related to the performance and efficiency requirements of the specific application. In general, the more the number of color cells, the more the histogram of color resolution is stronger. However, the large number of bin color histograms not only increases the computational burden, it also does not facilitate indexing in large image libraries. And for some applications, the use of very fine color space partitioning methods does not necessarily improve the search results, especially for those applications that do not tolerate errors in related images. Another way to effectively reduce the number of histogram bins is to use only those bins with the largest values (that is, the highest number of pixels) to construct the image features, because the bin representing the primary color can express the color of most of the pixels in the image. Experimental results show that this method does not reduce the retrieval effect of color histogram. In fact, due to the neglect of those smaller bins, the color histogram is less sensitive to noise and can sometimes make the search more effective.

(2) Color set
Color Histogram method is a global color feature extraction and matching method, can not distinguish the local color information. A color set is an approximation of a color histogram that first transforms an image from an RGB color space into a visually balanced color space (such as the HSV space) and quantifies the color space into several handles. Then, the image is divided into several regions by the color auto-segmentation technique, and each region is indexed by a color component of the quantization color space, thus expressing the image as a binary color index set. Comparing the distance between different image color sets and the spatial relationship of color regions in image matching

(3) Color moment

Color is one of the most important content of color image, which is widely used in image retrieval. However, when extracting color features from images, many algorithms first need to quantify the image. Quantitative processing can lead to false detection, and the resulting image feature dimension is high, which is not conducive to retrieval. Stricker and 0reng0 put forward the method of color moment [1], the color moment is a simple and effective method of color feature representation, There are first-order moments (mean, mean), second-order moments (Variance, viarance) and third-moment (slope, skewness), etc., because the color information is mainly distributed in the low-order moment, so with the first-order moment, second and third-order moment is sufficient to express the color distribution of the image, the color moment has been proved to effectively represent the color distribution in the image, therefore, the image of the color of a total of only 9 components (3 color components, 3 low-order moment It's very concise compared to other color features. The advantages of this method are: There is no need to quantify the color space, the feature vector dimension is low, but it is found that the retrieval efficiency of the method is relatively low, so it is often used to filter the image to narrow the scope of retrieval.

(4) Color aggregation vector

For the disadvantage that the color histogram and the color moment cannot express the spatial position of the image color, pass proposes the color aggregation vector of the image (color
Coherence vector). It is an evolution of the color histogram, whose core idea is to divide the pixels belonging to each bin of the histogram into two parts: if the area of the contiguous area occupied by some pixels in the bin is greater than the given threshold, the pixels within that region are aggregated pixels, otherwise they are non-aggregated pixels. Assuming that like and βi represent the number of aggregated and non-aggregated pixels in the bin of the histogram, respectively, the color aggregation vectors of the image can be expressed as < (α1, β1),
(α2, β2), ..., (αn, βn) >. and <α1 + β1, α2 + β2, ..., αn +βn > is the color histogram of the image. Because of the spatial information of the color distribution, the color aggregation vector can achieve better results than the color histogram.

(5) Color correlation diagram

Color Correlogram is another way of expressing the color distribution of an image. This feature not only depicts the percentage of pixels in a given color, but also reflects the spatial correlation between different color pairs. Experiments show that the color correlation graph has higher retrieval efficiency than color histogram and color aggregation vector, especially the image with consistent query spatial relationship.

If you take into account the correlation between any color, the color correlation graph becomes very complex and large (space complexity is O (n2d)). A simplified variant is the color auto-correlation graph (Auto-correlogram), which only examines spatial relationships between pixels with the same color, thus reducing the complexity of the space to O (Nd).

Reference: http://blog.csdn.net/langyuewu/article/details/4144139

http://blog.csdn.net/ts_zxc/article/details/20059827

Color Feature Extraction