Fusion of Visual and Thermal Face Recognition Techniques: A ...

More documents

Recommendations

Info

1.3 Thermal Face Recognition Face recognition in the thermal infrared domain has received relatively little attention in the literature in comparison with recognition in visible-spectrum imagery. Identifying faces from different imaging modalities, in particular from infrared imagery becomes an area of growing interest. Despite the success of automatic face recognition techniques in many practical applications, recognition based only on the visual spectrum has difficulties performing consistently under uncontrolled operating environments. Performance of visual face recognition is sensitive to variations in illumination conditions [33]. The performance degrades significantly when the lighting is dim or when it is not uniformly illuminating the face. Even when a face is well lit, differences in the angle of view can affect manual or automatic locating of feature points. Shadows, glint, makeup, and disguises can cause greater errors in locating the feature points and deriving relative distances. Thermal infrared images represent the heat patterns emitted from an object. Since the veins and tissue structure of a face is unique, the infrared images are unique. Thermal IR imagery is independent of ambient illumination since the human face and body is an emitter of thermal energy. The passive nature of the thermal infrared systems lowers their complexity and increases their reliability. The human face and body maintain a constant average temperature of about 36 degrees providing a consistent thermal signature. This is in contrast to the difficulties of face segmentation in the visible spectrum due to physical diversity coupled with lighting, color, and shadow effects. At low resolution, IR images give good results for face recognition. Thermal face recognition is useful under all lighting conditions including total darkness when the subject is wearing a disguise. Disguised face detection is of particular 9
interest in high-end security applications. Disguises are meant to cheat the human eye. Various disguise materials and methods that have been developed over the years are impossible or very difficult to be detected in the visible spectrum. Two methods for altering the facial characteristics of a person: artificial materials, e.g., fake nose, make-up, wig, artificial eyelashes, and surgical alteration. Visual identification of individuals with disguises or makeup is almost impossible without prior knowledge. The facial appearance of a person changes substantially through use of simplistic disguise such as fake nose, wig, or make-up. The individual may alter his/her facial appearance via plastic surgery. Both of these issues are critical for the employment of face recognition systems in high security applications. The thermal infrared spectrum enables us to detect disguises under low contrast lighting. Symptoms such as alertness and anxiety can be used as a biometric which is difficult to conceal as redistribution of blood flow in blood vessels causes abrupt changes in the local skin temperature. Radiometric calibration is an important step for thermal facial imagery. Thermal IR cameras can be radiometrically calibrated using blackbody radiation. Radiometric calibration achieves a direct relationship between the gray value response at a pixel and the thermal emission. This standardizes all thermal data even if taken under different environments. This relationship is called responsivity. The grayvalue response of thermal IR pixels is linear with respect to the amount of incident thermal radiation. The slope of this responsivity line is called the gain and the y-intercept is the offset. Images of a blackbody radiator covering the entire field of view are taken at two known temperatures, and the gains and offsets are computed using the radiant flux for a blackbody at a given temperature [34][35]. 10
Page 1 and 2: Fusion of Visual and Thermal Face R
Page 3 and 4: Abstract This paper describes how f
Page 5 and 6: List of Tables Table 1: Datasets...
Page 7 and 8: Figure 27: Example images of same i
Page 9 and 10: heat emitted by objects. The use of
Page 11 and 12: illumination source can reduce ligh
Page 13 and 14: identification recognition systems
Page 15: The LFA constructs a family of loca
Page 19 and 20: covers any area, which deals with u
Page 21 and 22: decision fusion schemes (average co
Page 23 and 24: sensors for experimentation and sta
Page 25 and 26: lights, we have two extra spot ligh
Page 27 and 28: taken with the eyeglasses off. For
Page 29 and 30: (a) (b) (c) Figure 6: A data fusion
Page 31 and 32: Edge detection methods such as Cann
Page 33 and 34: (a) (b) (c) Figure 9: Average therm
Page 35 and 36: 2 2 F(A,T)= AT = ax + bxy + cy + dx
Page 37 and 38: ignored to achieve better performan
Page 39 and 40: detection algorithm as described ab
Page 41 and 42: Eyeglasses (b) No eyeglasses (c) Fi
Page 43 and 44: Table 3: Performance for eyeglass d
Page 45 and 46: 4.1 Feature Based Face Recognition
Page 47 and 48: In addition to feature based method
Page 49 and 50: Figure 24: Accuracy results based o
Page 51 and 52: other factors such as compression a
Page 53 and 54: Figure 29: Pose test summary Table
Page 55 and 56: ‘Medium’ as not ambient but sti
Page 57 and 58: Based on the identification experim
Page 59 and 60: Figure 33: Decision fusion example
Page 61 and 62: in performance in this case. This i
Page 63 and 64: Figure 36: Performance comparison w
Page 65 and 66: Figure 39: Performance comparison o
Page 67 and 68:
average eye template in thermal ima
Page 69 and 70:
[14] P. S. Penev and J. J. Atick,
Page 71 and 72:
[30] P. J. Phillips, “Support Vec
Page 73 and 74:
[47] http://www.equinoxsensors.com/
Page 75:
Vita Jingu Heo was born in Daegu, K
show all

Fusion of Visual and Thermal Face Recognition Techniques: A ...

Create successful ePaper yourself

Delete template?

Save as template?