Detecting and Locating Human Eyes in Face Images Based on Progressive Thresholding Reporter: Kai-Lin Yang Date:2012/01/06 Authors: IEEE International Conference on Robotics and Biomimetics,

Outline Introduction Determination criterion of eye location Detecting and locating eyes in grayscale still images Experiments Detecting eyes using skin color segmentation Conclusions Comments 2

Introduction(1/3) This paper presents a novel algorithm for automatic detection and localization of human eyes in grayscale or color still images with complex background. First of all, a determination criterion of eye location is established by the priori knowledge of geometrical facial features. Secondly, a range of threshold values that would separate eye blocks from others in a segmented face image is estimated. 3

Introduction(2/3) Thirdly, with the progressive increase of the threshold by an appropriate step in that range, once two eye blocks appear from the segmented image, they will be detected by the determination criterion of eye location. To avoid the background interference, skin color segmentation can be applied in order to enhance the accuracy of eye detection. The localization of human eyes in face images is a fundamental step in the process of face recognition, works on color or grayscale still images (256 gray levels). 4

Introduction(3/3) A common method used for locating eyes is the deformable templates. 可變型樣板 (deformable templates) ◦ 任意形狀模型 : 不限制搜尋目標之幾何結構，凡符合限制的，都能被 找出。用於明顯可見的特徵。例如直線跟邊界。 ◦ 參數化模型 : 用於固定幾何結構之目標，因為結構固定所以適合用來 描述以及建造外觀，例如圓型跟矩形。 5

Determination criterion of eye location(1/5) To detecting and locating human eyes automatically and efficiently, suppose that face images are taken under the following reasonable conditions: ◦ Only one face appears in each image, ◦ The face takes up 15% to 50% of the whole pixels in the image. Wearing glasses is allowed, but no reflection of light and no black frame of glasses are expected because the reflection and black frame of glasses could seriously affect the results of detecting and locating eyes. 6

Determination criterion of eye location(2/5) 7

Determination criterion of eye location(3/5) The connected components (black pixels) in the segmented face image are called a block. To locate eyes from an appropriately segmented face image, a determination criterion of eye location needs to be established by the priori knowledge of geometrical facial features as follows: 8

Determination criterion of eye location(4/5) The distance between the geometrical centers of the two eye blocks should be within a certain range of pixel. There are no other blocks in a certain area under each eye. The vertical distance difference between the geometrical centers of the two eye blocks is not more than a certain number of pixels. 9

Determination criterion of eye location(5/5) The size (the pixel number) in each eye block is limited in a certain range. There is no other block between the two eye blocks. The proportion of height to length in the rectangular bounding box around each eye block is limited in a certain range. Any block connected with or very close to the four edges of face images is not an eye block. 10

Detecting and locating eyes in grayscale still images(1/6) We suppose that the optimal threshold value is in a range from T 0 to T max. When images are taken in a fixed background and an unchanged illumination condition, the range (T 0, T max ) can be limited in an even smaller range. 11

Detecting and locating eyes in grayscale still images(2/6) With the progressive increase of the threshold value by a small step T step in the range from T 0 to T max, we can see that in a segmented face image the size of the existing blocks will expand, some existing blocks will merge into one block, and some new blocks will emerge in the segmented face image 12

Detecting and locating eyes in grayscale still images(3/6) 13

Detecting and locating eyes in grayscale still images(4/6) 14

Detecting and locating eyes in grayscale still images(5/6) 15

Detecting and locating eyes in grayscale still images(6/6) 16 During the process, if finding r ≥ 0.5, the detected two eyes are usually true. If finding r < 0.5, the process will continue until finding r ≥ 0.5.

Experiments(1/1) 17 The average runtime is about 1 second per image and the correct localization rate is close to 98%.

Detecting eyes using skin color segmentation(1/5) 18 Compared to the skin color models commonly used in the current face detection algorithms, the skin color model we used is not for detecting face but for reducing the range of detecting eyes in order to enhance the accuracy of eye detection.

Detecting eyes using skin color segmentation(2/5) 19

Detecting eyes using skin color segmentation(3/5) 20 A simple rectangular bounding box around the region is taken for a skin color space D skin, which is used to segment the face image I c, giving a binary image, called a mask image I mask

Detecting eyes using skin color segmentation(4/5) 21

Detecting eyes using skin color segmentation(5/5) 22

Conclusions By automatic search for the optimal threshold value, a novel algorithm is developed in this paper for locating human eyes based on the determination criterion of eye location and the similarity measure of two eyes. 23

Comments 我覺得這篇偵測眼睛的方法讓人非常容易理解，不 過也有他不足的地方，就是如果拿不是人臉的圖片 來偵測，出現錯誤警告的機率太低，他還是會硬去 抓出類似眼睛的出來，所以可以從這邊下去改進他 的偵測錯誤。並且此方法眼睛的大小的門檻值是固 定的。 24