TWI514290B - Hand detection method and image procressing apparatus - Google Patents

Description

Hand detection method and image processing device

The present invention relates to a detection method and an image processing apparatus, and more particularly to a hand detection method and an image processing apparatus.

With the continuous advancement of technology, the development of human-machine interface has been changing with various advanced technologies and innovative applications. The use of gestures to identify human-machine interfaces to manipulate various devices has received increasing attention from the public.

Traditional solutions often require the use of auxiliary tools such as sensors on the user's fingers. This can increase the accuracy of hand detection, but it also increases the burden on the user. Another preferred way is to directly analyze the movement of the user's hand in an image processing manner, thereby manipulating the device. In the past, gesture detection and recognition methods required the establishment of a training database to detect the hand, such as the Haar-like feature, the Markov chain, or the like. The way of conversion removes the background, leaving only areas with similar skin tones, and then hand recognition through edge detection or convex hull.

However, the foregoing comparison method requires a large number of databases to be established and a long comparison time. Therefore, it is necessary to perform identification processing with a device with high processing performance, which is not easy to implement in a low-cost electronic consumer product. In addition, the way of skin color recognition is easily affected by the light source and similar skin color areas around the hand and the detection distance is limited. On the other hand, in order to avoid the influence of the edge information of the background, most of the methods using edge detection need to operate in a clean background, which does not conform to the actual use situation.

In view of the above, the present invention provides a hand detection method and an image processing apparatus that can efficiently detect a feature of a hand in a low-cost digital implementation.

The hand detection method provided by the present invention is applicable to an image processing apparatus, and includes the following steps: receiving an input image; performing edge detection processing on the input image to generate an edge image; and performing binarization processing on the edge image to Generating a binarized edge image, wherein the binarized edge image includes a plurality of edge pixels having a first pixel value and a plurality of non-edge pixels having a second pixel value; the edge painting of the binarized edge image In the prime, search for the starting pixel and the end pixel; determine whether the pixel value of the lower region between the starting pixel and the ending pixel conforms to the dynamic skin color interval associated with the input image; and when the starting pixel and the ending pixel are drawn When the pixel value of the lower region between the elements matches the dynamic skin color interval, it is determined whether the upper region between the start pixel and the end pixel has a plurality of finger features, thereby confirming the presence of the hand.

The image processing device of the present invention further includes a storage unit and one or more processing units, wherein the processing unit is coupled to the storage unit. Storage unit for recording A plurality of modules, and the processing unit is configured to access and execute the modules recorded in the storage unit. The module includes an image receiving module, an edge detecting module, a binarization module, a searching module, a skin color determining module, and a hand feature determining module. The image receiving module is configured to receive an input image. The edge detection module performs edge detection processing on the input image to generate an edge image. The binarization module is configured to perform binarization processing on the edge image to generate a binarized edge image, wherein the binarized edge image includes a plurality of edge pixels having a first pixel value and a plurality of second pixel values Non-edge pixels. The search module is configured to search for the starting pixel and the end pixel from the edge pixels of the edge image. The skin color judging module is configured to determine whether the pixel value of the lower region between the starting pixel and the end pixel meets the dynamic skin color interval associated with the input image. When the skin color judgment module determines that the pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval, the hand feature determination module is configured to determine the upper portion between the start pixel and the end pixel Whether the area has the characteristics of a plurality of fingers is used to confirm the presence of the hand.

Based on the above, the hand detection method and the image processing device of the present invention first perform edge detection on a single gradient of the input image to generate an edge image, and then set the edge binarization threshold value through the result of the dynamic skin color filter program. Then, the edge image is binarized to generate a binarized edge image. Then, the starting pixel and the ending pixel can be searched from the binarized edge image, and whether the lower region conforms to the dynamic skin color interval associated with the input image, and then three peaks and valleys are searched from the upper region to determine whether there is The finger feature is used to confirm the presence of the hand and to initially locate the position of the hand. No need to train database comparisons and pure skin tone In the case of identification, the hand detection method and device provided by the invention can not only achieve hand detection, but also achieve the performance of instant image or video processing, and is applied to low-cost consumer electronic products, and enhanced. The applicability of the invention in practical applications.

The above described features and advantages of the invention will be apparent from the following description.

100‧‧‧Image processing device

102‧‧‧ storage unit

104‧‧‧Processing unit

110‧‧‧Image receiving module

120‧‧‧Edge Detection Module

130‧‧‧ Binarization Module

140‧‧‧Search Module

150‧‧‧ skin color judgment module

160‧‧‧Hand feature judgment module

S202~S212‧‧‧Hand flow detection method

S _b , E _b , S _s , E _s ‧ ‧ pixels

F _d, B ‧‧‧ from

400‧‧‧Histogram

FIG. 1 is a block diagram of an image processing apparatus according to an embodiment of the invention.

2 is a flow chart of a method for detecting a hand according to an embodiment of the invention.

3(a)-3(b) illustrate binarized edge images in accordance with an embodiment of the present invention.

4 is a histogram of the number of edge pixels relative to a pixel column, in accordance with an embodiment of the present invention.

The components of the present invention will be described in detail in the following description in conjunction with the accompanying drawings. These examples are only a part of the invention and do not disclose all of the embodiments of the invention. Rather, these embodiments are merely examples of devices and methods within the scope of the patent application of the present invention.

1 is a block diagram of an image processing apparatus according to an embodiment of the present invention, but is for convenience of description and is not intended to limit the present invention. First of all FIG. 1 first introduces all components and configuration relationships of the image processing apparatus, and detailed functions will be disclosed together with FIG.

Referring to FIG. 1 , the image processing apparatus 100 of the present embodiment can perform hand detection on an input image to locate a hand position from the input image. The image processing apparatus 100 can be an electronic device having an image processing function such as a personal computer, a notebook computer, a tablet computer, a digital camera, a smart phone, or a scanner, and the present invention is not limited thereto. The image processing apparatus 100 includes a storage unit 102 and one or more processing units 104, the functions of which are described below.

The storage unit 102 is, for example, any type of fixed or removable random access memory (RAM), read-only memory (ROM), flash memory (Flash memory), hard A disc or other similar device or combination of these devices is used to record a plurality of modules executable by processing unit 104 that can be loaded into processing unit 104 for hand detection of the input image.

The processing unit 104 is, for example, a central processing unit (CPU), or other programmable general purpose or special purpose microprocessor (Microprocessor), digital signal processor (DSP), programmable Controllers, Application Specific Integrated Circuits (ASICs), Programmable Logic Devices (PLDs), or other similar devices or combinations of these devices. The processing unit 104 is coupled to the storage unit 102, which can access and execute the modules recorded in the storage unit 102.

The module includes an image receiving module 110, an edge detecting module 120, a binarization module 130, a search module 140, a skin color determining module 150, and a hand feature determining module 160. These modules are, for example, computer programs that can be loaded into the processing unit 104 to perform hand detection on the input image. Hereinafter, the detailed steps of the image processing apparatus 100 performing the image contrast enhancement processing will be described with reference to the embodiments.

2 is a flow chart of a method for detecting a hand according to an embodiment of the invention. The method of the present embodiment is applicable to the image processing apparatus 100 of FIG. 1, and the detailed steps of the hand detection method of the present invention are described below in conjunction with the components in the image processing apparatus 100. In the embodiment, the image processing device 100 is coupled to an image capturing device having an imaging lens and a charge coupled device (CCD) image sensor as an example for description. In another embodiment, the image capturing device may also be built in the image processing device 100, and the present invention is not limited thereto. Therefore, before the hand detection method is performed, the storage unit 102 of the image processing apparatus 100 has stored the video stream containing a plurality of images captured from the image capturing device in advance.

Referring to FIG. 2, first, the image receiving module 110 receives an input image (step S202). In detail, the input image includes a plurality of input pixels, and the input pixels are arranged in a matrix in a row and column manner. In this embodiment, the input image is a color image conforming to the YCbCr format commonly used to describe a digital image signal, and thus each input pixel includes a Y pixel value, a Cb pixel value, and a Cr pixel value. However, in other embodiments, the input image may also be an image located in another color gamut, and the present invention is not limited thereto. Here, the Y pixel value, the Cb pixel value, and the Cr pixel value are defined as the "component value" of the input pixel, wherein the Y pixel value is defined as the "luminance value", and The Cb pixel value and the Cr pixel value are defined as "color information".

Then, the edge detection module 120 performs edge detection processing on the input image to generate an edge image (step S204). Generally, when performing image edge detection, it is often necessary to simultaneously calculate the horizontal and vertical gradients. In addition, the light source or background tends to cause the edges of the image to be blurred, and the image is similarly reduced to cause edge blurring. Since the present invention is directed to the detection of the hand, and the vertical features of the hand are more obvious, in order to detect the edge features at a long distance, the edge detection module 120 in this embodiment will draw according to each input. The component value of the element is the vertical edge detection of the input image using a vertical Sobel filter.

In detail, in this embodiment, the edge detection module 120 first captures the component values of each input pixel of the input image, that is, the Y pixel value, the Cb pixel value, and the Cr pixel value. Then, the edge detection module 120 performs vertical edge detection processing on the component values of each input pixel according to Equations (1) to (3) to generate component derivative values: Where Y ( i , j ), Cr ( i , j ), and Cb ( i , j ) are the Y pixel value, the Cb pixel value, and the Cr pixel value of the input pixel whose coordinates are located at ( i , j ) in the input image, respectively. (*) is a convolution operator, For the vertical Sobel filter, EY ( i , j ), ECr ( i , j ), and ECb ( i , j ) correspond to the Y pixel value, the Cb pixel value, and the Cr pixel value component derivative value, respectively.

In addition, in order to detect a hand feature located at a long distance, for each input pixel, the edge detection module 120 will sum the component derivative values to generate an edge detection value. The purpose of summing the component derivative values here is to increase the information of the edge intensity, which can be expressed by equation (4): E ( i , j ) = EY ( i , j ) + ECb ( i , j ) + ECr ( i, j) equation (4) where E (i, j) is located at coordinate (i, j) of the input pixel corresponding to the edge detection value. Then, the edge detection module 120 uses the edge detection value E ( i , j ) to generate an edge image.

Next, the binarization module 130 performs binarization processing on the edge image to generate a binarized edge image (step S206). In other words, the binarization module 130 will binarize the edge image with a binarized threshold value a to generate a binarized edge image. In this embodiment, it can be expressed by equation (5): Where TE ( i , j ) is the pixel value of the coordinates ( i , j ) in the binarized edge image. When TE ( i , j ) is the first pixel value, it represents a pixel with edge features, which is defined as “edge pixel”; when TE ( i , j ) is the second pixel value, it represents A pixel that does not have an edge feature is defined herein as a "non-edge pixel." In an embodiment, the first pixel value may be 255 and the second pixel value may be zero.

It is worth noting that in order to preserve the gradient of the hand in the edge image and effectively filter the background information, the binning threshold α is related to whether the edge feature of the hand is obvious, and whether the noise is too much to be recognized or Misjudgment and other circumstances. When the value of α is lower, the edge information of the edge image is more obvious, but the easier it is to connect to other edge regions of the non-hand, the more background noise is, which is not conducive to subsequent analysis; otherwise, when the value of α is higher, Some of the edge source information will be filtered out, and the edge features of the hand may also be discontinuous. In this embodiment, the image processing apparatus 100 further includes a dynamic skin color filtering module (not shown), which can first perform a dynamic skin color filtering process according to the input image to adaptively adjust the binary value for different shooting environments. The threshold value is α.

In particular, the dynamic skin tone filter program filters out non-skinned areas in the input image. It must be noted here that the present invention does not judge the shape of the hand by the skin color, so that even if the subject is affected by the light source or the background has a similar skin color, the hand features of the skin color region are broken or incomplete, and Does not affect the results of the hand detection method. First, the dynamic skin color filter module can construct a Cb dynamic skin color interval MinCr < Cr < MaxCr and a Cr dynamic skin color interval MinCb < Cb < MaxCb according to the input color information (Cb pixel value and Cr pixel value) of the input image. MinCb , MaxCb , MinCr, and MaxCr are the minimum and maximum values of the Cb dynamic skin color interval and the minimum and maximum values of the Cr dynamic skin color interval, respectively. AvgCr and AvgCr are the Cb pixel values of the input pixel and the average of the Cr pixel values, respectively. Value, and satisfy equations (6)~(9): MinCr = AvgCr equation (6)

MaxCr = AvgCr +40 Equation (7)

Ma.xCb = AvgCb × 1.1 Equation (8)

MinCb = MaxCb -40 Equation (9) Next, the dynamic skin tone filtering module can determine the input pixel according to whether the Cb pixel value of the input pixel and the Cr pixel value simultaneously meet the Cb dynamic skin color interval and the Cr dynamic skin color interval. Skin color pixels or non-skin pigments.

In this embodiment, the binarization module 130 estimates the luminance values of the front and rear backgrounds in the input screen through the result of the dynamic skin color filtering program, and further sets the threshold value α of the binarization, so the threshold of binarization is used here. The value α is called the "edge binarization threshold". The binarization module 130 sets the edge binarization threshold value by three values of the reference image brightness average value, the skin color brightness average value, and the non-skin tone brightness average value. Here, the image brightness average is the average value of the luminance values (Y pixel values) of all the input pixels; the skin color brightness average is the average value of the skin color pixels; the non-skin brightness average is the non-skin pixel The average of the brightness values. In this embodiment, the binarization module 130 may first find the reference image brightness average value, the skin color brightness average value, and the non-skin skin brightness average value Minmin , the intermediate value Midlum, and the maximum value Maxlum , according to the equation ( 10) Obtain the edge binarization threshold α to make the edge features more obvious: However, in the present embodiment, if the α calculated by the equation (11) is less than 65, the binarization module 130 still sets the edge binarization threshold α to 65 to avoid the result of the binarization. There was too much noise.

After the binarization module 130 generates the binarized edge image, the image processing device 100 searches for the hand by using the characteristics of the finger and the palm of the hand when the palm of the palm is compared with the image capturing device in the posture of "5". s position. First, the search module 140 searches for the starting pixel and the end pixel from the edge pixels of the binarized edge image (step S208). In detail, the search module 140 can find the first edge pixel and the second edge pixel from top to bottom and from left to right from the binarized edge image, and determine the first edge pixel and the second edge. Whether the distance between the edge pixels is greater than the palm width threshold value and whether there is no other edge pixel between the first edge pixel and the second edge pixel. The reason why the palm width threshold is set here is that if the distance between the first edge pixel and the second edge pixel is too small, the corresponding gradient will be unclear. In addition, since the vertical information of the palm of the palm is small, the edge pixels generated after the binarization processing are small. Therefore, when the search module 140 determines that the distance between the first edge pixel and the second edge pixel is greater than the palm width threshold and that there is no other edge pixel between the first edge pixel and the second edge pixel, the search The module 140 sets the first edge pixel and the second edge pixel as the starting pixel and the ending pixel respectively; otherwise, the searching module 140 continues from the binarized edge image to the right, and then looks down. New first edge pixels and new second edge pixels. Taking the binarized edge image 300 of FIG. 3 as an example, the starting pixel and the ending pixel searched by the search module 140 may be S _b and E _b , respectively, where the distance between S _b and E _b is B.

Next, the skin color determination module 150 determines whether the pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval associated with the input image. (Step S210). In the present embodiment, the lower area here may be a square area formed below the start pixel and the end pixel, but the present invention is not limited thereto. When the skin color determination module 150 determines that the component value of the input pixel corresponding to the lower region conforms to the dynamic skin color interval, it means that the input pixel corresponding to the lower region has a large amount of skin color components, that is, it is highly likely to be a palm region. Therefore, the hand feature determination module 160 can perform further feature determination in the subsequent step S212. When the skin color determination module 150 determines that the component values of the input pixels corresponding to the lower region do not conform to the dynamic skin color interval, that is, the start pixel and the end pixel are not features of the palm, so the image processing apparatus 100 re Step S208 is executed to enable the search module 140 to re-search for a new starting pixel and a new ending pixel.

On the other hand, when the skin color determination module 150 determines that the pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval, the hand feature determination module 160 determines the start pixel and the end pixel. Whether or not the upper region has a feature of a plurality of fingers is used to confirm the presence of the hand (step S212). In other words, when the skin color determination module 150 determines that the lower region between the start pixel and the end pixel is highly likely to be a feature of the palm, the hand feature determination module 160 can further determine the start pixel and the end point. Whether the upper area between the elements is a feature of the finger. Since the height of the finger of a general human does not exceed 1.1 times the width of the palm, the hand feature determination module 160 can take the starting pixel S _b and the end point by taking the binarized edge image 300 of FIG. 3(b) as an example. The distance between the pixels E _b and the upper F _d is to find the pixel S _s and the pixel E _s , where F _d =1.1× B . In the present embodiment, the region 302 formed between the start pixel S _b , the end pixel E _b , the pixel S _s and the pixel E _s is the aforementioned upper region.

Next, the hand feature determination module 160 will determine whether the upper region has the characteristics of the finger. In detail, the hand feature determination module 160 determines whether each column of the pixel in the upper region is a peak pixel column or a valley pixel column and whether the position of the column corresponds to the finger. Fingertip features or finger joint features of a finger. In this embodiment, the hand feature determination module 160 will first look in the upper region for the presence of three peak-valley combinations. The combination of peaks and valleys here has a certain distance between two adjacent peak pixel columns, and there is a valley pixel column between them. As can be seen from FIG. 3(b), if the hand feature determination module 160 finds the features of the three fingertips, the judgment can be completed as long as the features of one finger seam are found on both sides of each fingertip. Since the thumb is far away in the state of the palm of the hand, and the height of the little finger is far from the length of the other fingers, it is not the judgment condition of the embodiment. In other words, when the peak feature determination module 160 finds the peak pixel column, it is determined whether the distance between two adjacent peak pixels is not greater than the width threshold and whether there is a valley pixel between them. . In this embodiment, the width threshold value may be 1/4 hand width. The following is a further explanation of the steps for finding a combination of three peaks and valleys.

First, the hand feature determination module 160 can be up from the bottom (ie, the line segment according to the upper region 302 in FIG. 3(b). To the line segment The direction) is judged. The hand feature determination module 160 can accumulate how many edge pixels are in each column of the horizontal projection of the upper region 302 and record it in the vector His [0: m ], where m is the width of the upper region 302. For convenience, the vector His [0: m ] can be represented by the histogram 400 of FIG. 4, wherein the horizontal axis of the histogram 400 is the index of the vector His [0: m ], that is, the number of the pixel column. The vertical axis of the histogram 400 is the number of edge pixels. The hand feature determination module 160 can look up from the vector His [0: m ] whether it has a maximum value greater than, for example, 0.8 times or more the height of the upper region 302. If not, the hand feature determination module 160 goes to the line segment. The direction continues to look upward; if it is and its maximum value is the value of the index in the vector His [0: m ], which is index k (that is, the maximum value is His [ k ]), the hand feature determination module 160 can learn from the vector. His [0: m] is greater than the acquired His [k] corresponding to the value of the index N ₁ × 0.8, and the recording medium P in the vector. It is assumed that in the present embodiment, the index corresponding to the larger value of N ₁ in the vector His [0: m ] is 2, 3, 6, and 9, and the vector P = (2, 3, 6, 9). Here, the index recorded by the vector P is the number of the aforementioned peak pixel column, that is, the feature associated with the fingertip. On the other hand, the hand feature judging module 160 takes an index corresponding to the value N ₂ smaller than, for example, His [ k ] × 0.5 from the vector His [0: m ], and records it in the vector V. It is assumed that in the present embodiment, the index corresponding to less than N ₂ in the vector His [0: m ] is 0, 1, 4, 5, 7, 8, 10, and 11, and the vector V = (0, 1, 4, 5,7,8,10,11). The index recorded in this vector V is the number of the aforementioned valley pixel column, that is, the feature associated with the finger seam.

The hand feature judging module 160 can determine whether the fingertip and the finger seam are actually present in the upper region 302 according to the vector P and the vector V , that is, whether the starting pixel S _b and the end pixel E _b are present three. Peak and valley combination. In detail, the hand feature determination module 160 can search for a value between two adjacent elements in the vector V from the vector P in groups of two adjacent elements. For example, 2 and 3 of the vector P = (2, 3, 6, 9) are continuous values, so there is no element of vector V between the two values. On the other hand, the values between 3 and 6 are 4 and 5, which exist in the vector V ; the values between 6 and 9 are 7 and 8, which are also present in the vector V.

Next, the hand feature determination module 160 determines whether the distance between each two adjacent peak pixel columns is not greater than the width threshold. In this example, since m = 11 and the hand width is 12, the width threshold will be set to 3, which is 1/4 hand width. Here, the hand feature determination module 160 determines that there is no more than 3 between 3 and 6 and between 6 and 9. When the distance between two adjacent peak pixels is not greater than the width threshold, the hand feature determination module 160 then determines two of the leftmost peak pixel column and the rightmost peak pixel column. Whether there is at least one valley element column on the side. If so, the hand feature determination module 160 can determine that the upper region 302 conforms to the three peak-to-valley combinations to confirm the presence of the hand. For example, in this embodiment, the hand feature determination module 160 can determine that the left side of the element 2 of the vector P = (2, 3, 6, 9) has the elements 0 and 1 of the vector V , and the vector P The right side of element 9 has elements 10 and 11 of vector V.

In an embodiment, after the hand feature determination module 160 confirms the presence of the hand in step S212, the hand can be positioned according to the position of the finger to binarize the edge image. That is to say, the peak pixel of the binarized edge image corresponds to the fingertip feature of the hand, and the valley pixel corresponds to the finger joint feature of the hand. In an embodiment, after the hand feature determination module 160 locates the hand, it can be marked, and the image processing apparatus 100 can re-execute step S208 to determine whether there is another hand feature in the input image. In addition, in an embodiment, after the hand feature determination module 160 confirms that the hand does not exist in step S212, the image processing apparatus 100 re-executes step S208 to continue searching for the hand feature.

In an embodiment, the image processing device 100 can track the features of subsequent opponents. For example, if the hand is kept facing upward, the search module 140 can narrow the space and time of the search because the movement range of the hand is limited, and the dynamic skin color filtering program does not need to be re-executed, and the hand feature determination module 160 You can reposition your opponent.

In summary, the hand detection method and the image processing device of the present invention first perform edge detection on a single gradient of the input image to generate an edge image, and then set edge binarization through the result of the dynamic skin color filter program. The threshold value, and then the edge image is binarized to produce a binarized edge image. Then, the starting pixel and the ending pixel can be searched from the binarized edge image, and whether the lower region conforms to the dynamic skin color interval associated with the input image, and then three peaks and valleys are searched from the upper region to determine whether there is The finger feature is used to confirm the presence of the hand and to initially locate the position of the hand. In the case of no training database comparison and pure skin color recognition, the hand detection method and device provided by the invention can not only achieve hand detection, but also achieve the performance of instant image or video processing, and use The applicability of the present invention in practical applications is enhanced on low cost consumer electronics.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any one of ordinary skill in the art can make some changes and refinements without departing from the spirit and scope of the present invention. The scope of the invention is defined by the scope of the appended claims.

S202~S212‧‧‧Hand flow detection method

Claims (10)

A method for detecting a hand is applied to an image processing device, comprising: receiving an input image; performing an edge detection process on the input image to generate an edge image; performing a binarization process on the edge image a binarized edge image, wherein the binarized edge image includes a plurality of edge pixels having a first pixel value and a plurality of non-edge pixels having a second pixel value; In the edge pixel of the edge image, searching for a starting pixel and a final pixel; determining whether a pixel value of a lower region between the starting pixel and the ending pixel conforms to a corresponding image of the input image a dynamic skin color interval; and when a pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval, determining an upper region between the start pixel and the end pixel Whether there are multiple finger features to confirm the presence of one hand. The hand detection method of claim 1, wherein the step of performing the edge detection processing on the input image to generate the edge image comprises: inputting a pixel for each input image: Taking a plurality of component values of each input pixel of the input image; performing a vertical edge detection process for each of the component values to generate a plurality of component derivative values; and accumulating the component derivative values for Generating an edge detection value; The edge image is generated according to the edge detection value of each of the input pixels. The hand detection method of claim 2, wherein the component values of each of the input pixels comprise a brightness value and at least one color information, the dynamic skin color interval being associated with the input pixel The color information, each of the input pixels includes a plurality of skin color pixels that meet the dynamic skin color interval and a plurality of non-skin pixels that do not meet the dynamic skin color interval, and the binarization processing is performed on the edge image. The step of generating the binarized edge image includes: calculating an image brightness average value, a skin skin brightness average value, and a non-skin skin brightness average value according to the brightness value of each of the input pixels, wherein the image brightness average And an average value of the brightness values of the input pixels, wherein the skin color brightness average is an average value of the brightness values of the skin color pixels, and the non-skin skin brightness average is the non-skin pixel An average value of the brightness values; an edge binarization threshold value is set according to the image brightness average value, the skin color brightness average value, and the non-skin skin brightness average value; and according to the edge The binarization threshold is performed for each pixel of the edge image to generate the binarized edge image. The hand detection method according to claim 1, wherein the step of searching for the starting pixel and the end pixel from the edge pixel comprises: determining a first of the edge pixels Whether a distance between a edge pixel and a second edge pixel is greater than a palm distance threshold, and the first edge pixel and the second edge pixel have no other edge pixels; and if Setting the first edge pixel and the second edge pixel respectively The starting pixel and the endpoint pixel. The hand detection method of claim 1, wherein when the pixel value of the lower region between the starting pixel and the end pixel meets the dynamic skin color interval, determining the starting painting Whether the upper region between the element and the endpoint pixel has the finger feature, the step of confirming the presence of the hand includes: determining whether there are three peak-to-valley combinations in the upper region, wherein the three peaks The valley combination includes a plurality of peak pixel columns corresponding to a fingertip feature and a plurality of valley pixel columns corresponding to a finger joint feature, the two adjacent peak points of the three peak valley combinations There is at least one of the valley elements between the pixels, and the distance between any two adjacent peaks of the three peaks and valleys is not greater than a width threshold, the three At least one of the two sides of the combination of peaks and valleys further includes at least one other of the valley elements; and if so, confirming the presence of the hand and positioning the location of the binarized edge image based on the finger feature The hand. An image processing apparatus includes: a storage unit that records a plurality of modules; and one or more processing units coupled to the storage unit to access and execute the module recorded in the storage unit, the module The group includes: an image receiving module that receives an input image; an edge detection module that performs an edge detection process on the input image to generate an edge image; and a binarization module for the edge image , performing a binarization process, Generating a binarized edge image, wherein the binarized edge image includes a plurality of edge pixels having a first pixel value and a plurality of non-edge pixels having a second pixel value; a search module, The edge pixel of the edge image searches for a starting pixel and a final pixel; and a skin color determining module determines whether the pixel value of a lower region between the starting pixel and the end pixel meets a dynamic skin color interval associated with the input image; and a hand feature determination module, wherein the skin color determination module determines that a pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval The hand feature determination module determines whether an upper region between the start pixel and the end pixel has a plurality of finger features, thereby confirming the presence of a hand. The image processing device of claim 6, wherein the edge detection module captures a plurality of component values of each input pixel of the input image for each input pixel of the input image, Performing a vertical edge detection process on each of the component values to generate a plurality of component derivative values, and accumulating the component derivative values to generate an edge detection value, and the edge detection module according to each The edge detection value of the input pixel generates the edge image. The image processing device of claim 7, wherein the component value of each of the input pixels comprises a brightness value and at least one color information, and the image processing device further comprises: a dynamic skin color filtering module, according to the color information of the input pixel, defining a dynamic skin color interval, and distinguishing the input pixels that meet the dynamic skin color interval into a plurality of skin color pixels and not meeting the dynamic skin color The input pixel of the interval is a plurality of non-skinning pixels, wherein the binarization module calculates an image brightness average value, a skin color brightness average value, and a non-skin color according to the brightness value of each of the input pixels. An average value of the brightness, wherein the average value of the brightness of the image is an average value of the brightness values of the input pixels, and the average value of the brightness of the skin color is an average value of the brightness values of the skin color pixels, and the average value of the brightness of the skin color For the average value of the brightness values of the non-skin pixels, the binarization module sets an edge binarization threshold according to the image brightness average value, the skin color brightness average value, and the non-skin skin brightness average value. And the binarization module performs the binarization processing on each pixel of the edge image according to the edge binarization threshold to generate the binarized edge image. The image processing device of claim 6, wherein the search module determines whether a distance between a first edge pixel and a second edge pixel in the edge pixel is greater than a palm distance threshold. a value, and the first edge pixel and the second edge pixel have no other edge pixels; and if so, the search module sets the first edge pixel and the second edge pixel respectively The starting pixel and the endpoint pixel. The image processing device of claim 6, wherein the skin color determination module determines that the pixel value of the lower region between the start pixel and the end pixel meets the dynamic skin color interval, the hand The feature determination module determines the upper portion Whether there are three peak-to-valley combinations in the region, wherein the three peak-to-valley combinations include a plurality of peak pixel columns corresponding to one fingertip feature and a plurality of valley pixel columns corresponding to a finger joint feature, There is at least one of the gluten pixels column between any two adjacent peak pixel columns of the three peak-to-valley combinations, and any two adjacent peaks of the three peak-valley combinations The distance between the pixels is not greater than a width threshold, and at least one of the two sides of the combination of the three peaks further includes at least another of the valley pixels, and if so, the hand feature determination module The presence of the hand is confirmed and the hand is positioned based on the finger feature at the location of the binarized edge image.