Proc. of Int. Conf. on Multimedia Processing, Communication& Info. Tech., MPCIT

Color Histogram Approach for Analysis of Psoriasis Skin Disease B.V.Dhandra1, Shridevi Soma2, Shweta Reddy2, Gururaj Mukarambi3 2

1,3 Department of Computer Science, Gulbarga University, Gulbarga Department of Computer Science and Engineering, P.D.A College of Engineering, Gulbarga Email : [email protected]

Abstract-Psoriasis is an immune-mediated skin disease that cannot be cured completely but statuesque can be maintained under medication. In this paper a system is proposed based on color features for analysis of psoriasis skin disease. There are four types of psoriasis viz. Guttate, Nail, Plaque and Pustular. The objective of the paper is to diagnose the type of psoriasis for treatment based on color histogram features. Feature extraction process is carried out by converting the RGB color space of an image to HSV color space, and H and S planes are used to represent color histogram by dividing each plane into 11 bins, where each bin represents the color count. A total of 121 features are considered by taking the product of bins from H and S plane, and these features are normalized by dividing number of pixels in the bin to the total number of pixels. For multiclass classification, the Support Vector Machine (SVM) with RBF kernel is used. The system was tested on a set of 180 medical images and the experimental result confirms the efficiency of the proposed method as 92.2% accuracy. Keywords-SVM, RGB, HSV, RBF I. INTRODUCTION

Now a days people of different age groups are suffering from skin diseases and lesions such as eczema, scalp ringworm, skin fungal, skin cancer of different intensity, diabetic ulcers, psoriasis symptoms etc. The above said diseases strike suddenly without warning and have been one among the major disease that has life risk for the past ten years. If skin diseases are not treated at earlier stage, then it may lead to complications in the body including spreading of the infection from one individual to the other. The skin diseases can beprevented by investigating the infected region at an early stage. The characteristic of the skin images are diversified, so that it is challenging job to devise an efficient and robust algorithm for automatic detection of the skin disease and its severity. Skin tone and skin color plays an important role in skin disease detection. Color and coarseness of skin are visually different. Automaticprocessing of such images for skin analysis requires quantitative discriminator to differentiate the diseases. So, the choice of color becomes an important component in skin disease detection. Hence, an attempt is made to propose a new system for the analysis of psoriasis skin disease based on the color features. One of the major issues in detection of a skin disease using skin color is how to choose a suitable color space. Numerous color models (RGB, CMY, HSI, (HSV), Normalized RGB, and YCbCr) are used for skin disease detection. In this paper we are using a HSV color space for extracting the features from the infected skin images. Color histogram is used for calculating the color feature vector of each image and these features are used for the classification. Support Vector Machine with RBF kernel is used for multiclass classification of the skin images. DOI: 03.AETS.2013.4.44 © Association of Computer Electronics and Electrical Engineers, 2013

The remainder of this paper is organized as follows, Section II contains the related literature on skin disease detection techniques. Data collection and methodology used for skin disease detection is presented in Section III. Section IV contains the color histogram for feature extraction. In section V experimental results and discussions are presented. Conclusion is the part of Section 6. II . LITERATURE REVIEW In the literature, large amount of techniques can be found with respect to the detection of skin diseases. Some of the related articles are discussed in this section. Anal Kumar Mittra et al.[1] proposed an automated system for recognizing disease conditions of human skin using texture feature. Disease conditions are studied by using Gray Level Co-occurrence Matrix. Multilayer perceptron (MLP) classifier is used to detect the diseases and they have obtained 96.6% accuracy for disease detection. S.Arivazhagan et al.[2] presented an automated system for recognizing human skin diseases using texture features. The texture features are extracted from the gray level run-length matrices and Minimum Distance Classifier is used to classify the type of human skin diseases and have obtained an accuracy of 92.72%, where as Anal Kumar Mittra et al. [1] have used same feature with MLP and obtained an accuracy of 96.6%. Alaa Yaseen Taqa et al. [3], developed a robust skin detection method that integrates both color and texture features. The Back-propagation neural network is used for classification. They found that their proposed skin detection method achieves a true positive rate of approximately 94.5% and a false positive rate of approximately 0.89%. Color, texture and shape features are integrated by Zhiwei Jiang et al. [4] for the detection of skin disease. A marker driven watershed transform is used to demonstrate the accuracy of 94.8%. C.Prema et al. [5] presented a hybrid approach for skin tone detection. The Hybrid approach integrates both Cheddad’s approach and the threshold value of Cr which is obtained from YCbCr. The Minimum Distance classifier is used for classification and obtained an accuracy of 93.81%. Radu Dobrescu et al.[6], have described a method for automatic detection of malignancy of skin lesions which is based on both local fractal features and texture features are derived from the medium co-occurrence matrices. Their texture features are derived from the average grey level co-occurrence matrix and fractal features from fractal dimension. In the box counting algorithm, the fractal dimension is considered for different binary thresholds, and finally they have considered a mean value of these dimensions. On the other hand, the algorithms for local fractal dimension and local connected fractal dimension assume the histogram plotting of these local dimensions in all selected region pixels. Their experimental results confirmed the efficiency of the method as 93.05% accuracy. J. S. Taur et al.(2006), proposed a method for the segmentation of color images using a multiresolution-based signature subspace classifier (MSSC) with application to psoriasis images. A MSSC was adopted to segment the images and to save the computational time. Since the variations of skin color of the psoriasis and normal regions for different patients are sometimes very large, the feature signatures obtained from one image may not work well for other images. In order to obtain accurate segmentation, the proper training regions for an image can be manually selected to reflect the variations. In their approach, the fuzzy texture spectrum and the two-dimensional fuzzy color histogram from the hue-saturation space are first adopted as the feature vector to locate homogeneous regions in the image. Then these regions are used to compute the signature matrices for the orthogonal subspace classifier to obtain a more accurate segmentation. Their method does not require any a priori knowledge about the signature abundances of the color and texture. The method is quantitatively evaluated by using a similarity function and compared with the well-known LS-SVM method by experimentation. Based on the results of evaluation and comparison, the proposed method is shown to be effective on the psoriasis vulgaris images with an accuracy of 94.61%. A skin disease diagnosis system was developed and tested by Nidhal K. Al Abbadi et. al. [2010]. In their study both color and texture features are used to give a better and more efficient recognition accuracy of skin diseases, but the accuracy of identification is not reported in their paper. They have used the feed forward neural network to classify the input image to be psoriasis infected or non psoriasis infected. H. Hashim(2013) [20] presents a study on recognition of psoriasis features via daubechies d8 wavelet technique. Transformation of 2D Discrete Wavelet Transform (DWT) algorithm for Daubechies D8 method at first level is used to obtain the coefficients of the approximations and details sub-images. For classification, statistical approach analysis is applied to identify significance difference between each groups of psoriasis in terms of mean and standard deviation parameter. 78

Building an image based system to automatically score psoriasis [21] developed by David Delgado gomez consist of two elements. A precise image acquisition system and a statistical procedure to extract the lesions from the image. Lesions are isolated by a linear mixture of two Gaussians obtained from the features. Gaussian corresponds to healthy and diseased distributions. The estimates of mean and variance makes it possible to identify the lesion via quadratic discriminant analysis. Skin diseases are one of the diseases that are wide spread. Recently there are many machine vision systems developed for skin disease like skin cancer, eczema, scalp ringworm, psoriasis etc. Psoriasis is a common, Psoriasis

Low

Medium

High

Guttate

Nail

Plaque

Pustular

Fig. 1 The skin database sample with psoriasis

chronic, relapsing, inflammatory skin disorder with a strong genetic basis. A diagnosis of psoriasis is usually based on the appearance of the skin. So the image processing techniques help in diagnosing the disease by extracting the features from the infected skin images. From the above literature it is clear that, the detection of skin disease is an important and challenging problem in general and analysis of psoriasis skin diseases in particular. In view of this we made an attempt to devise an algorithm which is simple and robust to identify four different types of psoriasis with three severity levels. III. DATA COLLECTION AND METHODOLOGY The Fig-1 shows input samples of psoriasis skin images that are considered in this paper. There are total 180 images of 4 types and each type of image is subdivided as low, medium, high severity. Table-1 presents the psoriasis image sample distribution with respect to type of psoriasis and severity. The database contains both real images which are captured from the digital camera of 8 mega pixel and standard images are taken from www.psoriasisspot.com and en.wikipedia.org. The following Fig-2 presents the flow of the proposed system. Here all the input images are captured from the digital camera with certain distance and some images are also taken from the standard database. The preprocessing techniques are applied to crop the input image into 64*64 size pixels, then the RGB image is converted to HSV color space. IV. FEATURE EXTRACTION A. Color Feature Color is one of the most important features that make possible the recognition of images by human. Color is a property that depends on the reflection of light to the eye and the processing of that information in the brain. Here color histogram is used for color feature extraction.

79

B. Color Histogram A color histogram is representation of the distribution of colors in an Image. The color histogram can be built for any kind of color space, although the term is more often used for three-dimensional spaces like RGB or HSV. TABLE I. DATA COLLECTED FOR PSORIASIS INFECTED SKIN IMAGES Sl.No.

Class

Total no. of images

1

Guttate Low

12

2

Guttate Medium

10

3

Guttate High

13

4

Nail Low

17

5

Nail Medium

13

6

Nail High

11

7

Plaque low

12

8

Plaque Medium

20

9

Plaque High

20

10

Pustular low

20

11

Pustular Medium

10

12

Pustular High

22

Total

I/P Images

180

Preproces sing

Feature extracti on

Knowle dgebase

SVM Training

Test Image

Preproces sing

Feature Extractio n

class

Knowledge base

Fig. 2 Flow of proposed algorithm

A color histogram represents the distribution of colors in an image, through a set of bins, where each histogram bin corresponds to a color in the quantized color space. A color histogram for a given image is represented by a vector: H = {H[0], H[1], H[2], H[3], … … … … H[i], … … … , H[n-1]} where i is color in bin and H[i] represents the number of pixels of color i in the image, and n is the total number of bins used in color histogram. Typically, each pixel in an image will be assigned to a bin of a color histogram. Accordingly in the color histogram of an image, the value of each bin gives the number of pixels that has the same corresponding color. In order to compare images of different sizes, color histograms should be normalized. The normalized color histogram H´ is given as: H´= {H´[0], H´[1], H´[2], H´[3], … … … … H´[i], … … … , H´[n-1]} 80

where H´[i]=H[i]/p, p is the total number of pixels in an input image C. Histogram Generation From the HSV Color Space The HSV color space in general, and the HSV color histogram in particular, plays an important role in image analysis. In this paper an RGB image is taken for the analysis. The RGB image is converted into HSV color space. The choice of considering the HSV image is attractive in theory as it is considered to be more suitable, since it separates the color components (HS) from the luminance component (V) and is less sensitive to illumination changes, so in next step we are extracting only H and S planes which represents color component of an image as shown in the Fig. 3.

Fig. 3 Flow chart of Feature extraction algorithm

After the H and S plane separation hue scale is divided into eleven bins, saturation scale is divided intoeleven bins. Each bin contains the percentage of pixels in the image that have corresponding H and S colors for that bin. By combining each of these bins, we get a total of 121 color features to represent a HSV color histogram. After extracting the features, normalization process is carried out to obtain the feature vector. The above process is carried out for the training images to obtain the color features and these features are dumped into the features file. These 121 features of each training images is used for classification. Algorithm :Identification of Psoriasis skin disease Input: Preprocessed Image Output: Diseased Skin Class and Severity Start Step-1: Read the image. Step-2: Convert RGB color space image into HSV color space. Step-3: Carry out color quantization using color histogram by assigning 11 levels each Hue and Saturation to present a quantized HSV space with 11*11=121 histogram bins. Step-4: Obtain the normalized histogram by taking the ratio of pixels in the bin to the total number of pixels. Step-5: Repeat step1 to step4 on all images of the database to get a normalized database. Step-6: Select the query image. Step-7: Generate the normalized histogram of a query images by applying step1-step4. Step-8: Classify the query image using SVM with RBF kernel value 2. Step-9: Retrieve result. Stop 81

V. EXPERIMENTAL RESULTS AND DISCUSSIONS Support Vector Machines are supervised learning models with associated learning algorithms that analyze data and recognize patterns, used for classification and regression analysis. For non linear separable case many kernel mapping functions are used. In this paper a Radial Basis Function is used. A. Radial Basis Function: K (xi , xj) = exp(-γ xi - xj 2) , γ > 0

(1)

Where, γ controls width of the RBF kernel. We are using one against all decomposition method for classification. Experiment is carried out for the psoriasis skin disease classification using Support Vector Machine. The database consists of 4 different classes such as guttate, nail, plaque, pustular psoriasis and each class further divided into 3 categories low, medium and high based on severity. In Each class 10 to 20 images with resolution of 64*64 pixels are considered. The database contains total of 180 images with different types and severity of the psoriasis skin disease. Features are extracted from each of 180 images in the database. Some images are taken from the digital camera and some from standard database images such as www.en.wikipedia.org and www.psoriasisspot.com. The Fig.4 shows the graph for comparison of the identification rate for the different types of psoriasis skin diseases with three severity levels. The appendixes given below represents the intermediate results of the experiment carried out for the analysis. TABLE II. SUMMARIZES THE NUMBER OF IMAGES CONSIDERED IN EACH CLASS WITH AN IDENTIFICATION ACCURACY RATE OF DISEASED SKIN IMAGES. Sl. no

1 2 3 4 5 6 7 8 9 10 11 12

CLASS Guttate Low Guttate Medium Guttate High Nail Low Nail Medium Nail High Plaque low Plaque Medium Plaque High Pustular low Pustular Medium Pustular High

Total no. of images

No. of correctly classified images

No. of incorre-ctly classified images

Classification Rate

12

12

00

100%

10

09

01

90%

13

13

00

100%

17

17

00

100%

13

13

00

100%

11 12

09 12

02 00

80% 100%

20

18

02

90%

20

20

00

100%

20

16

04

100%

10

08

02

80%

22

18

04

Average Accuracy Rate

80% 92.2%

VI. CONCLUSION In this paper, color histogram based features are used to analyze and classify the psoriasis infected skin images in order to take the diagnostic measures. On one hand this would be useful for dermatologists to reduce diagnostic errors, while on the other hand it can serve as the initial test bed for patients in rural areas where there is a dearth of good medical professionals. A support Vector Machine with RBF kernel is used for the classification of images. The experimental results gave the encouraging results in an initial attempt for identification of psoriasis infected skin images. Classification of psoriasis skin diseases and their severity will be carried out. The future work is to reduce the number of features and to increase the classification accuracy.

82

100 80 60

High

40

Medium

20

Low

0 Guttate

Nail

Plaque Pustular

Fig. 4 Analysis of Psoriasis infected Skin images

[21] Devid Delgado et al, “Building an image-based system to automatically score psoriasis”, 13th Scandinavian conference on Image analysis, 2003 APPENDIXES

Fig. 5 Menu Window

Fig. 8 Testing Image

Fig. 6 Sample of a Trained image

Fig. 9 Matching image retrieved for selected query image

Fig. 7 Histogram features of input images

Fig 10. Skin class with severity

REFERENCES [1] Anal Kumar Mittra & Dr.Ranjan Parekh ,“Automated Detection Of Skin Diseases Using Texture Features” JadavpurUniversity ,Kolkata India (6 June 2011). [2] S.Arivazhagan et al, “Skin Disease Classification By Extracting Indepentend Components” T.N, India (10 Oct 2012).

83

[3] Alaa Yaseen Taqa & Hamid A.Jalab, “Constructing Reliable Skin Detector Based On Combining Texture And Color Features” University of Mosul (2 March 2011). [4] Zhiwei Jiang et al, “Skin Detection Using Color, Texture and Space Information” Wuhan University, China(august 2007). [5] C.Prema & D.Manimegalai, “A Novel Skin Tone Detection Using Hybrid Approach By New Color Space” T.N, India (7 May 2012). [6] Radu Dobrescu et al, “Medical Image Classification For Skin Cancer Diagnosis Based On Combined Texture And Fractal Analysis” University Of Bucharest, Romania (7 July 2010). [7] J.S.Taur et al, “Segmentation Of Psoriasis Vulgaris Image Using Multiresolution-Based Orthogonal Subsapace Techniques” IEEE 2006. [8] Nidhal k. Al Abbadi et al, “Psoriasis Detection Using Skin Color and Texture Features” University of Kufa,Iraq(2010). [9] Peter Howarth & Stefan Ruger, “Evaluation Of Texture Feature For Content-Based Image Retrieval” London (2004). [10] Handaru Jati & Dhanapal Durai Dominic, “Human Skin Detection Using Defined Skin Region” University of Petronas (IEEE 2008). [11] Chelsia Amy Doukim et al, “Combining Neural Networks For Skin Detection” University of Malaysia, Japan (SIPIJ 2010). [12] M.Babu Rao et al, “Content Based Image Retrieval Using Dominant Color, Texture And Shape” A.P ,India (4 April 2011). [13] Sagar soman et al, “Content Based Image Retrieval Using Advanced Color And Texture Features” Mumbai, India(ICCIA 2011). [14] Alexandru Caliman et al, “Color Fractal Dimension For Psoriasis Image Analysis” University of Poitiers, France 2011. [15] Umasankar Kandaswamy et al, “Robust Color Texture Feature Under Varying Illumination Conditions” IEEE 2012. [16] Dr.Ranjan Parekh, “Using Texture Analysis For Medical Diagnosis” Jadavpur University ,Kolkata India (IEEE 2012). [17] Jing Zhang et al, “ROI Based Natural Image Retrival Using Color And Texture Feature” GyeongSang National University, Korea. [18] Assdollah Shahbahrami et al, “Comparison Between Color And Texture Feature For Image Retrieval” University of Guilan ,Iran. [19] Vladimir Vezhnevets et al, “A Survey On Pixel Based Skin Color Detection Techniques” Moscow State University, Russia [20] H. Hashim et al, “Recognition of Psoriasis features via Daubechies d8 wavelet technique” Universiti Teknologi MARA, 40450 Shah Alam, Selangor, Malaysia

84