An Effective Segmentation Pattern Using Multi-class Independent Component Analysis on High Quality Color Texture Images

An efficient segmentation pattern proposes to improve the efficiency of segmentation through MultiClass Independent Component InfoMax Analysis (MICIA) on multi-class high-quality color images. To attain richer segmentation of color, textures with minimal computation time, MICIA combines the watershed cuts principle and Minimal Spanning Forest method. The higher quality texture image is segmented by using the watershed cuts principle. Watershed cuts principle in MICIA is associated with regional minima of the map to handle multi-class poorly defined boundary images. Independent Component Analysis (ICA) is based on InfoMax which achieves richer segmentation of color textures with maximum likelihood function. ICA is based on InfoMax. It handles multiclass texture images. Because of the ICA maximum likelihood ensures higher independence on segmentation cuts.This produces an effective segmentation which can be used to improve the appearance of the high-quality images. To prove the efficiency, the experiment is conducted on factors such as sub-pixel accuracy rate on segmenting, multi-class image segmentation time and true positive rate.


INTRODUCTION
The effective segmentation pattern is still a challenging task, though many segmentation methods and approaches have been presented in the recent years.With the increasingly high-quality texture images, application of image segmentation has received greater attention ever before.
Realizing its need, HAIRIS (Gonçalves et al., 2011) has presented a method to perform segmentation on a pair of images on the basis of relaxation parameter.HAIRIS has also applied histogram modes via the area among the objects.It is an efficient method for image registration and is proven as a robust statistical-based method for effective object matching.HAIRIS with different spectral content is applied only to single-band images.But it is not implemented towards multi-band image segmentation and it also consumes more computational time during single-band segmentation stage.
A novel mathematical and algorithmic model is called as UIS (McCann et al., 2014) which has designed to perform unsupervised image segmentation.Flexible segmentation framework with the unsupervised method is, even though well suited for the segmenting wide range of multi-class poorly defined boundary images, it consumes more time on simply segmenting the colors from multi-class images.Model-based adaptation framework (Ecabert et al., 2011) includes multi-class feature and left atrium.The proximal part of the pulmonary veins, including coronary sinus evaluate various sizes of heart chambers in a significant and consistent manner.Generalized Hough Transformation is applied to multi-class images to minimize the computation time.Even though the efficiency is improved, it is not efficient in addressing other modalities.
For effective recognition of objects, image segmentation plays the significant task of detailed understanding and analysis of images.Two novel methods (Ghamisi et al., 2012), namely Fractional-Order Darwinian Particle Swarm Optimization (FODPSO) and Darwinian Particle Swarm Optimization (DPSO) have been developed for measuring the optimal threshold level for a specific image given as input.Even though more stable with less CPU time is ensured, the computational complexity of the algorithm remains low.To improve the complexity related to computation, a method is called as Minimum Spanning Tree (MST-based graph) (Zhong et al., 2011) has been presented using K-means.With minimal parameters, the effect is achieved.It does not suit for high dimensional datasets.
One of the critical tasks related to medical image processing is automatic image segmentation.It might be applied for effective diagnosis of disease and provisioning of treatments accordingly.Image Segmentation Automated Oracle (ISAO) (Frounchi et al., 2011) is presented to build an oracle which forms the basis for automatic verification and validation for image segmentations.The method not only reduces the use of resources, but also, efficient classification is made between consistent and inconsistent segmentation pairs with respect to each segmentation pair.Even though the method seems to be promising, the number of iterations required to perform the entire process remains unaddressed.
A maximum likelihood approach (Chen et al., 2011) has efficiently performed an automatically tuned process by using the Cramer-Rao Bound method.With this, the registration accuracy is improved in addition to optimally fused performance.
One of the most popular methods for sensing remote objects over the past few years is geographic object-based image analysis, which is used for evaluating and measuring high spatial resolution images.A new automated model for parameterizing multi-scale image segmentation is presented in Drăguţ et al. (2014).Automation and objectivity are increased and the scale to which it can be applied is relatively less addressed.In order to improve the level of scalability (Chen et al., 2010), a joint segmentation and registration technique is designed by using data fusion hierarchical structures.
For effective interpretation of dynamic scene representation, segmentation of structure-and-motion is of high significance for performing effective segmentation.A combinatorial framework (Thakoor et al., 2010) has been designed to keep in mind the optimization of the cost function.It combines three factors, namely, maximum likelihood of hypotheses, cost involved in clustering and distribution of an outlier in a uniform manner to minimize computational complexity.The drawback of the framework highly relied on the initial value of hypothesis.An improved threshold-based segmentation (Abdullaha et al., 2012) has been presented to the partition of the natural images in a clear manner and reduces the complexity involvement during computation applying an inverse technique.
The main focus is made on improving the efficiency of segmentation using the watershed cuts principle and independent component analysis based on informatics method.With the application of the Watershed cuts principle, richer segmentation of color textures with minimal computation time has been attained.The Watershed cut principle in MICIA is associated with regional minima of map effectively and handles multi-class poorly defined boundary images by using dilation and erosion of points on twodimensional multi-class images.
Independent Component Analysis is based on InfoMax which achieves richer segmentation of color, textures with maximum likelihood function.As a result of ICA based on InfoMax handling multi-class texture images, the Maximum likelihood ensures higher independence on segmentation cuts.Finally, the Minimal Spanning Forest method evaluates the image at minimum timing intervals.High-quality texture image reduces the computation time on multi-class images and improves the sub-pixel accuracy rate on segmenting.

Different clustering methods have been designed
for performing image segmentation.The artificial bee colony algorithm (Ouadfel and Meshoul, 2012) has been applied to improve the performance of the image being segmented.In addition to that, a new mutation strategy is also introduced to improve the process involvement during exploitation.Effectiveness and efficiency of segmentation (Mignotte, 2012) are achieved at the cost of time.An image segmentation framework using nonlinear dimensionality reduction is introduced.To minimize the complexity with respect to time, a contour is integrated with texture cues and performed subsequent segmentation.
Retrieval and segmentation of images using a range of colors, modularity of texture and shape are significant research areas of study with widened scope.A new framework (Jain et al., 2012) is designed to integrate all the three metrics namely color, texture and information regarding shape in order to perform effective segmentation and to enhance the retrieval rate at a lesser amount of time using the dominant color feature.Even though the precision is improved, the complexity increases with the increasing number of features applied.
To reduce the complexity involved during segmentation (Salah et al., 2011), multi-region graph cut image partitioning is applied to evaluate the deviation from the original images by improving the flexibility of the model involved.However, dimensionality remains unaddressed by minimizing the retrieval rate.A hierarchical approach (Jayaraman et al., 2011) is designed and combines both iris color and texture to improve the retrieval rate of the iris images.An algorithm called Speeded-up Robust Features algorithm is used to improve the performance of iris recognition with respect to the hit rate and high penetration rate.
The discussed techniques presents the design of multi-class independent component Informax analysis for improving the efficiency of segmentation on highquality color texture images.

METHODS AND MATERIALS
This section provides a brief overview of design considerations involved in multi-class independent The watershed principle is efficiently constructed to provide the symbolic representation of the segmentation process by using MICIA based segmentation.With the application of mathematical minimum spanning forest method, the operations involved during the image segmentation reduce the computational time with effective segmentation results.With this, the minimum spanning forest method eventually segments the images with minimal computation time by traveling through the shortest distance to segment the multi-class images.In addition, the watershed cut principle satisfies optimality property in MICIA based segmentation using the MSF method.The minimum spanning forest represented on the color texture image is depicted in Fig. 2. The overall process of MICIA method is depicted in Fig. 3 using the architecture diagram.
As illustrated in Fig. 3, MICIA method initially extracts the multi-class color images from Corel Image Feature database.In order to perform efficient segmentation on multi-class color images, the watershed principle is used to segment different color textures for easy classification process during the upcoming work.The process of segmentation is carried out in MICIA method using the minimum spanning forest method in order to reduce the computational time.Followed by this, the segmentation process uses the multi-class independent component analysis using the Infomax method.The application of multi-class independent component analysis minimizes the class dependencies between two components.Here dependency refers to the regions of the observed vector that has dependent mixture color textures.The independent component used in MICIA method effectually reduces the dependency level (i.e., the rate of dependency of color textures in a given color texture database) and, in turn, improves the likelihood function.
A Maximum likelihood function is used in a MICIA method for richer segmentation attainment.The brief note on watershed principle using minimum spanning forest method and multi-class independent component analysis using Infomax is explained in the forthcoming sections.

Water shed principle using MSF method:
The watershed principle is applied on two-dimensional images I (x, y) with the region 'R' segmentation, then the watershed gradient on Image 'I' is expressed as: , In (1), 'D' denotes the color dilation of points on two-dimensional, multi-class images whereas, 'E' denotes the erosion of color points (i.e.,) that is highly unrelated to the defined segmented structure and is therefore removed using the dividing units where ' ' represents the gray value of image ' '.Further, consider that if , <0 is considered as minimal zones of color while regions with , >0 is considered as the maximal zone of color texture on two dimensional multi-class images.Then, the watershed principle is applied to a MICIA method in order to successfully identify, dilation or an erosion of color occurs on twodimensional multi-class images 'I'.
In an MICIA method based on the minimum spanning forest (Zhong et al., 2011) creates an optimality of watersheds.Minimum spanning forests relative to sub-pixels of 'P' induce a unique pixel cut by using the watershed principle.The main result of segmentation by using the MICIA method with minimum spanning forest is to identify the relative minimum distance for mapping the pixel points.In fact, MICIA method derives the minimum spanning tree computations using distinct weights.As a result, the weight (where weight represents the sum of vertices and edges) on each pixel point is likely used on the watershed cut property: Let 'a' and 'b' denote the two-pixel points on the image 'I' and identify whether the pixel point 'a' is a minimum spanning forest pixel to perform the segmentation process or not.If the condition does not get satisfied, then the 'b' pixel point is checked through its corresponding weight points.Each pixel on watershed principle with MSF is explained through the algorithmic procedure.The above algorithmic step describes the minimum spanning forest procedure for segmenting the regions by traveling through shortest distance.The edges of the poorly bounded images are also segmented effectively in MICIA method.Regional minimal result with minimal computation time is produced by MICIA method.The weight of each pixel point helps to plot effective segmentation regions.
Multi-class independent component analysis: Once, the richer segmentation of color, texture is obtained using watersheds cut principle, the goal of the multiclass independent component analysis is to widely separate the mixed color textures.Assuming a twodimensional vector image c (t) = [ , ,  … ], such that the components (i.e.,) regions are mutually independent on time 't'.The vector c (t) corresponds to 'M' independent scalar valued points.The multi-class probability distribution function of the vector is measured as the product of marginal independent distributions: In (3), 'P' denotes the probability distribution function using MICIA method and the product operations is effectually carried out at all independently scaled value points 'M'.With this, different regions (i.e.,) color textures are segmented effectively.
Infomax method: An input pixel vector points in MICIA method is observed at each time point t, such that the regions of the observed vector are no longer dependent.The regions of c (t) are designed in such a way that if one source pixel points are normally distributed, then it is possible to extract the remaining pixel points to be segmented from image 'I'.With this, the multi-class normal distribution point is formalized as: The component independence is checked through (4) where the regions ' , , … ' on twodimensional vectors are mutually independent.The above equation in Infomax method for independent component analysis is simplified as: From above, the mutual information in MICIA method is always positive and also equals to zero only when the components are independent on mapping the segmented regions.

Maximum likelihood function:
Maximum likelihood function in MICIA method achieves richer segmentation processing signal with lesser dependency rate.MICIA picks up the higher order values of the normal distribution and performs the redundancy reduction:

€
: , , … 0 (6) In order to form richer segmentation, with the application of maximum argument likelihood, ' ' in (6) is the identified regions.Maximum likelihood based on the journal "IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 20, NO. 5, MAY 2011" with the titled " A Maximum Likelihood Approach to Joint Image Registration and Fusion" by the authors "Chen et al. (2011)", enables to separate independent components in MICIA method input data with normally distributed function value.Maximum likelihood provides the unified result in richer segmentation without dependency rate.

RESULTS AND DISCUSSION
Multi-Class Independent Component Info Max Analysis (MICIA) based segmentation method is used to improve the color texture segmentation with higher efficiency rate.High-quality image segmentation using MICIA method is implemented in MATLAB.Segmentation processing uses the Corel Image Features Data Set from UCI repository for experimental work.Corel Image Features Data Set holds 68,040 photo images from a mixture of categories such as high quality and low-quality images.
Each set of features in Corel Image Features Data Set is stored in a separate file and each line corresponds to a single image used for segmenting the color textures.The initial value of the corel image features Data set is the image ID and the succeeding standards are the feature vector (e.g., Color textures) of the image.The similar image has the equivalent ID in all files, but the image ID is not same as the image filename in Corel Image Features Data Set.Co where, is the maximum allowed value in twodimensional images with and which are the samples to be considered for left and right of the sub-pixel?The efficiency of color, texture segmentation ( ) is measured by using ( 4) and ( 5) that evaluates the rate of efficiency of the regions, ' , , … ' with that of the multi-class distribution point, : (8) The multi-class image segmentation time measures the time was taken to perform multi-class image segmentation using the watershed gradient on two dimensions, , : , The True Positive Rate (TPR) in MICIA method uses a maximum likelihood function that performs redundancy reduction.There is the total number of true positive pixel points, is divided by the actual positive pixel points : Multi-Class Independent Component InfoMax Analysis (MICIA) on multi-class high-quality color images are compared against the existing Histogrambased Image Segmentation (HAIRIS) (Gonçalves et al., 2011) and Unsupervised Image Segmentation (UIS) (McCann et al., 2014).The evaluation value is given below with the help of the table and the graph describes the MICIA on multi-class high quality a color image which improves the sub pixel accuracy rate.Figure 4 evaluates the sub-pixel accuracy rate measured in terms of percentage (%) achieved with different number of sub-pixel ranging from 100 to 700 and a comparison is made with the two existing schemes, namely, Histogram-based Image Segmentation (HAIRIS) (Gonçalves et al., 2011) and Unsupervised Image Segmentation (UIS) (McCann et al., 2014).
The color, texture segmentation efficiency of our work MICIA with the existing two schemes, namely Histogram-based Image Segmentation (HAIRIS) (Gonçalves et al., 2011) and Unsupervised Image Segmentation (UIS) (McCann et al., 2014) is provided in Fig. 5.
Figure 6 describes the multi-class image segmentation time based on the number of regions taken into consideration for experimental purpose in the range of c 1 to c 7 .The number of regions is computed by (4).By using the application of the watershed cuts principle richer segmentation of color, textures are obtained with minimal computation time, which improves the multi-class image segmentation time is achieved by 25-37% when compared to HAIRIS (Gonçalves et al., 2011).Watershed cut principle in   MICIA is associated with regional minima of the map to handle multi-class poorly defined boundary images improving the multi-class image segmentation time by 62-89% when compared to UIS (McCann et al., 2014).More accurately the influence of the true positive rate with respect to the number of pixel points is listed in Table 4 and the comparison is made with two other existing schemes.It can also be seen that the true positive rate increases with the increase in the number of pixel points.
More accurately the influence of the true positive rate with respect to the number of pixel points is provided in Fig. 7 and comparison is made with two other existing schemes.It can also be seen that the true positive rate increases with the increase in the number of pixel points.
Figure 7 describes the true positive rate based on the number of pixel points in the range of 200 to 1400.From the Fig. 7 it is illustrative that the true positive rate using the proposed MICIA is than the two other existing methods, namely, HAIRIS (Gonçalves et al., 2011) and UIS (McCann et al., 2014) respectively.This is because the independent component used in MICIA method effectually reduces the dependency level and in turn, improves the likelihood function.A Maximum likelihood function is used in the MICIA method for richer segmentation attainment resulting in improved true positive rate by 7-53% and 14-80% when compared to HAIRIS (Gonçalves et al., 2011) and UIS (McCann et al., 2014) respectively.
Figure 8 given shows the qualitative analysis of sub pixel accuracy and color text segmentation efficiency using the proposed MICIA method and comparison made with existing HAIRIS and UIS respectively.
From the Fig. 8, the image provided as input is in Fig. 8a.Sub Pixel accuracy obtained using Multi-Class Independent Component InfoMax Analysis (MICIA) on multi-class high quality color images are given in Fig. 8b and the comparison is made with two other methods Histogram-based Image Segmentation (HAIRIS).Unsupervised Image Segmentation (UIS) is provided in Fig. 8c and d.Color Text Segmentation efficiency measured using the proposed MICIA is provided in Fig. 8f and comparison made with HAIRIS and UIS is provided in Fig. 8g and h respectively.

CONCLUSION
Effective segmentation of multi-class high quality color images has become the key for image processing.To achieve richer segmentation of color, texture with minimal computation time and improve the level of color, texture segmentation efficiency with the relatively lesser amount of multi-class segmentation time remains the objective of this study.In this study, the performance effects of high quality color, texture images and handle multi-class poorly defined boundary images had been investigated by proposing a method, Multi-Class Independent Component InfoMax Analysis (MICIA).The MICIA method based on Watershed cuts principle and Independent Component Analysis based on InfoMax provides an efficient means of richer segmentation of color textures with maximum likelihood function.At first, the use of watershed cut principle minimum spanning forest method that measures the computation time on multi-class images are studied and proposed to use color dilation and

Fig. 3 :
Fig. 3: Architecture diagram of MICIA methodFigure2depicts the procedure involved in the design of a Minimum Spanning Forest (MSF) method on multi-class high-quality color texture images.The spanning tree is constructed with minimal computation time on high quality images.The Watershed principle with pixel cuts using the MSF method rapidly performs the segmentation process.The overall process of MICIA method is depicted in Fig.3using the architecture diagram.As illustrated in Fig.3, MICIA method initially extracts the multi-class color images from Corel Image Feature database.In order to perform efficient segmentation on multi-class color images, the watershed principle is used to segment different color textures for easy classification process during the upcoming work.The process of segmentation is carried out in MICIA method using the minimum spanning forest method in order to reduce the computational time.Followed by this, the segmentation process uses the multi-class independent component analysis using the Infomax method.The application of multi-class independent component analysis minimizes the class dependencies between two components.Here dependency refers to the regions of the observed vector that has dependent mixture color textures.The independent component used in MICIA method effectually reduces the dependency level (i.e., the rate of dependency of color textures in a given color texture database) and, in turn, improves the likelihood function.
Let us assume that 'a' and 'b' be the pixel point on image 'I' to perform segmentation Step 2: Initially, Watershed Principle is applied to compute the dilation and erosion of color, texture image Step 3: Repeat Step 4: Compute edge of the region to easily plot color, texture space on two-dimensional images 'I' Step 5: Check to see that if two pixel points, 'a' and 'b' is within the cycle of the plotted points Step 6: If pixel points weight (a>b) then Step 6.1: Pixel point 'a' is chosen for segmentation process Step 7: Else Step 7.1: Pixel point 'b' is chosen for segmentation process Step 8: Until all pixel points in the cycle of 'I' use MSF procedure Output: MSF reduces the computational time of segmenting multi-class 'I' End

Fig. 4 :
Fig. 4: Measure of sub pixel accuracy rate with respect to number of sub pixel

Fig. 6 :
Fig. 6: Measure of multi-class image segmentation time Fig. 8: (a): Input image; (b): Sub pixel accuracy, using MICIA; (c): Sub pixel accuracy, using HAIRIS; (d): Sub pixel accuracy, using UIS; (e): Input image; (f): Color text segmentation efficiency using MICIA; (g): Color text segmentation efficiency using HAIRIS; (h): Color text segmentation efficiency using UIS erosion points on two dimensional, multi-class images for measuring relative minimum distance for mapping the pixel points in MICIA method.Secondly, the multiclass independent component analysis is performed using Infomax method to assess multi-class normal distribution point with varying regions to maximize the true positive rate using Corel Image Features Data Set extracted from the UCI repository.The experiment conducted using Corel Image Features Data Set shows that the MICIA method achieves up to 17.25% improvement in color texture segmentation efficiency compared to the existing methods.

Table 1 :
Tabulation for sub-pixel accuracy rate

Table 2 :
Tabulation for color, texture segmentation efficiency

Table 4 :
Tabulation for true positive rate