What Is Histogram Image Processing?

Image histograms are widely used in various areas of image processing due to their low computational cost and many advantages such as image translation, rotation, and scaling invariance. In particular, threshold segmentation of grayscale images, color-based image retrieval, and images classification.

Image histograms are less computationally expensive and have images

For image segmentation

The gray histogram, color histogram, visual histogram, and image segmentation, image retrieval, and image classification applications built on these histograms are specifically studied here. The main research work is as follows:

1. Aiming at the problems of poor adaptability of one-dimensional threshold segmentation algorithm, susceptibility to noise interference, and high complexity of two-dimensional threshold segmentation algorithm, this paper proposes an integrated threshold segmentation algorithm based on region division of gray histogram. The algorithm uses pixel gray level and field mean to form a two-dimensional space. On this two-dimensional space, the method of region division is used to construct a one-dimensional histogram, and then it integrates three classic segmentation algorithms: minimum error, maximum entropy, and maximum inter-class variance. , Construct a new threshold selection method, and finally use the acquired threshold segmentation algorithm to have a more powerful adaptation and robust noise immunity: compared with the two-dimensional algorithm, the adaptive noise type is more, and the algorithm's calculation complexity is small a lot of.

2. The traditional color distance measurement usually uses the simplest European distance measurement. However, in the HSV color space, because the components contribute differently to color, this simple color distance measurement cannot be transplanted here. On space. Aiming at this problem, this paper proposes a parameterized HSV color space distance to distinguish different components by the difference of parameters. Then manually label the relative size of the distance between the color pairs, and use the pair-wise-based learning method to train the labeled data, and finally learn the parameters in the distance formula to obtain a parameterized color space distance formula suitable for the HSV space.

3. In view of the four problems of losing spatial location information, it is proposed that the color histogram has four problems: high computational dimension, susceptibility to changes in brightness, lack of correlation between similar colors, and loss of spatial location information. A clustering-based Spatial color histogram method. This method first performs k-means clustering on images. Then the spatial color histogram of the spatial position distribution is statistically considered on the clustering graph. At the same time, the parameterized distance formula of the HSV color space at the training place is trained by the manual annotation data method, and the similarity matching algorithm of the spatial color histogram is given above. Experiments show that the method is simple to implement and reflects the image color and other characteristics. The image retrieval effect based on the proposed algorithm is better than the traditional method. In addition, the method is highly adaptable and can be labeled with corresponding data for training as needed to meet diverse subjective color similarity needs.

Bag-of-words is a classic method for image classification. The core is how to select visual words and how to count visual word histograms. This paper proposes an image classification algorithm based on visual word histogram. This method first uses the visual attention mechanism to apply to the classified images to obtain the saliency map of the images. Then extract a variety of image features such as color and shape, and use the saliency map to construct weights. Generate a dictionary, count the visual word histograms of each image, and then use the L1 regularized logistic regression method to filter the features. Finally, use the SVM classifier to make classification decisions, and finally obtain the image classification results.