Abstract—This paper describes an automatic detection and recognition system of leukocytes on a given microscopic image. The developed system detects the locations of leukocytes from a blood cell image. After the automatic detection, the system classifies each leukocyte in one of the five categories (neutrophils, eosinophils, basophils, lymphocytes, and monocytes). The system processes an input image with the Scale Invariant Feature Transform (SIFT) algorithm for leukocyte detection. Meanwhile, two different recognition methods, i.e. the Euclidean distance and the Co-occurrence matrix methods are applied for automatic recognition. The combination of detection and recognition approaches provides the optimal recognition accuracies for almost all leukocyte types. Index Terms—Leukocyte detection, leukocyte recognition, microscopic image, scale invariant feature transform. I. INTRODUCTION Blood is a bodily fluid that delivers nutrients and oxygen to cells. The analysis of blood cells can be used to detect blood disorder or to determine the presence of infectious diseases in human body. In order to identify the hematopoietic system disorders, hematologists need to perform the blood cells identification and counting for every blood elements, such as the erythrocytes (red cells), leukocytes (white cells), and platelets [1]. Since the task is very tedious and really time consuming, an automaticblood detection, recognition, and counting system is necessary and helpful. Several researchers have proposed various methods todetect and recognize the blood cells, such as the work by Markiewicz using the Support Vector Machine method [2], Colunga with EMalgorithm [3], and Neural Network-based classifiers [4], [5]. However, the detection and recognition systems have not been tested for blood cells that were influenced by rotation or illumination effects after the segmentation process. In this paper, an automatic leukocyte detection system that can detect the white blood cell locations from microscopic images is developed. The proposed system works based on the Scale Invariant Feature Transform (SIFT) method. First, the system detects the white blood cells locations using the SIFT method, then the system crops the images which contains the region of interest. After the automatic detection Manuscript received August 7, 2014; revised November 13, 2014. This work was supported in part by the Indonesian Directorate General of Higher Education under Hibah Penelitian Unggulan Perguruan Tinggi Tahun Anggaran 2014. Lina A. Chris, B. Mulyawan, and A. B. Dharmawan are with the Computer Science Department, Faculty of Information Technology, Tarumanagara University, Jl. Letjen. S. Parman 1, Jakarta 11440, Indonesia (e-mail: lina@untar.ac.id). and cropping, the system will recognize the leukocyte type using two different recognition methods: 1) the distance based recognition system using the Euclidean distance method, and 2) the color based recognition system using the co-occurrence matrix method. The remainder of this paper is organized as follows. In Section II, the proposed leukocyte detection system based on SIFT algorithmis explained. Section III presents the leukocyte recognition system, while Section IV describes the experimental setup and results. Finally, the conclusion is presented in Section V. II. LEUKOCYTE DETECTION SYSTEM In the proposed system, the Scale Invariant Feature Transform (SIFT) method is applied to detect the leukocytes from the captured microscopic images. The SIFT algorithm, developed by Lowe [6]-[8] is an algorithm for image features generation which are invariant to image translation, scaling, rotation and partially invariant to illumination changes and affine projection [9]. The steps for defining the SIFT image features are as follows: 1) Scale space construction, 2) Keypoint localization, 3) Orientation assignment, 4) Keypoint descriptor. First, the system creates a scale space from the input imagesby calculating the Difference of Gaussian (DoG) using the Gaussian kernel. This step is necessary as an input image may consist unncessary details for detection or recognition processes. Therefore it is important to identify locations and scales that contains only the region of interest from an image. The first step for detecting locations that are invariant to scale changes is by constructing a continuous function of scale, known as the scale space. The scale space of an image is defined as a function, L(x, y, σ), that is produced from the convolution of a variable-scale Gaussian, G(x, y, σ), with an input image I(x, y) (,,) (,,) (,)   Lxy σ Gxy σ Ixy (1) where * is the convolution operation in x and y, and 2 2 2 2 ( )2 (, , ) (1 2 )      x y Gxy σ e (2) To build the DoG pyramid, the input image is convolved iteratively with the Gaussian kernel. The last convolved image is down-sampled in each image direction by factor of 2, and the convolving process is repeated [9]. Each collection of images of the same size are then build together the so-called Gaussian pyramid, which is represented by a 3D function defined in Eq. 1. The DoG pyramid is computed from the difference of each two nearby images in Gaussian pyramid. Fig. 1 shows the constructed scale space images. A Leukocyte Detection System Using Scale Invariant Feature Transform Method Lina Arlends Chris, Bagus Mulyawan, and Agus Budi Dharmawan International Journal of Computer Theory and Engineering, Vol. 8, No. 1, February 2016 69 DOI: 10.7763/IJCTE.2016.V8.1022