In the texture segmentation with the various texture features, the existing histogram clustering algorithms have drawbacks in the memory storage requirement and computational time because it utilizes the multidimensional histogram in the feature space.
In this thesis, An algorithm for the texture segmentation using the conditional 1-D histogram is proposed. The algorithm consists of the following steps.
First, d feature data files are prepared for the inputs of this proposed algorithm. The texture features, i.e mean, variance, correlation(horizontal, diagonal, vertical) etc, are extracted by the sum and difference histogram method.
Second step, the 1-D histogram is constructed with the first texture feature data(mean), and then the threshold values of the clusters are determined by the peak-valley detection method. Using these thresholds, we can segment the original image by the first texture feature only.
Third step, the 1-D histograms of each cluster determined by the first texture feature(mean) are constructed with the second texture feature(variance), and then split the clusters by the peak-valley detection method.
These histograms are so called the conditional 1-D histograms considering the first texture feature(mean). Now, we can split the original image by the first (mean) and second(variance) texture features.
Next, in the same way, repeat the above third step to the d th feature (horizontal, diagonal and vertical correlation).
In the result of the simulations on the artificial and natural textured images, we can obtain the desirable segmented images. With the suggested algorithm, due to the conditional 1-D histogram instead of the multidimensional histogram, the memory storage requirement and processing time can be reduced.