CN103077406A - Image recognition method based on grid marks - Google Patents

Abstract

The invention discloses an image recognition method based on grid marks, which belongs to the technical field of computer image recognition. It takes the internal connectivity of the image as the main identification feature, and inputs it into the computer to complete the identification of the internal texture of the digital image. The method of obtaining the eigenvalues is: dividing the image into several grids of the same size, scanning and marking the state of each grid, and obtaining the connected features between each grid, so as to realize the classification and recognition of the image. The image recognition method provided by the invention improves the recognition efficiency under the premise of ensuring the recognition accuracy, is especially suitable for sorting and inspection of objects with different internal patterns recognized by a computer in large batches, and has a good application effect.

Description

An Image Recognition Method Based on Grid Marking

technical field

The invention relates to an image recognition method, in particular to an image recognition method based on grid marks, and belongs to the technical field of computer pattern recognition.

Background technique

The application of computer to image recognition has been widely used in various fields. The classification of input images by computer can help people to classify objects quickly and effectively, thus greatly saving the cost of manual screening. For example, in recent years, image recognition has been increasingly used in the identification, sorting and grading of tiny objects, such as the sorting of screws in industry and the quality inspection of grain in agriculture. The application of image recognition in the field of object sorting has greatly reduced labor costs and doubled the efficiency of sorting, and minimized the misselection rate. However, the accuracy of image recognition has always been a problem that plagues people.

In the existing technology, effective image recognition algorithms include: contour perimeter area ratio based on object boundary, Fourier series method, pixel proportion method based on pixel point distribution, and connectivity discrimination method based on image texture, etc. In the application process of image recognition, most algorithms have shortcomings such as low efficiency of image recognition algorithms, insensitivity to changes in image texture, and great influence by image noise. This situation limits the popularization and application of automatic recognition. For example: the recognition method based on the object boundary can effectively distinguish the outline of the object, but it cannot recognize the surface pattern of the image itself; the method based on the pixel point distribution can distinguish the color distribution and light and shade of the object image, but it also Unable to distinguish texture effectively.

At present, the commonly used connectivity calculation algorithms mainly include: region growing method, tracking algorithm, boundary tracking method, based on row (column) scanning algorithm, etc. However, none of the above methods can guarantee the reliability of the calculation results when the image is noisy.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art and provide an accurate and efficient recognition method suitable for large batches of images with large differences in internal patterns.

In order to achieve this goal, the technical solution adopted in the present invention is: provide a kind of image recognition method based on grid mark, it is characterized in that distinguishing image with the degree of connectivity of image, comprises the following steps:

1. Divide the image to be identified into several equal grids according to the granularity requirements, and pre-set the size of the grid. The minimum initial value of the grid size is 3×3 pixels;

2. Scan each grid point in the image and identify the pixel distribution parameters in each grid, including: ① byte-type row and column boundary variable status; ② index-type connected state variable mark;

3. Merge the blank areas in the grid and record their connectivity status;

4. Divide the non-blank grid into smaller grids again. If the grid size is still greater than the minimum value, return to step 2;

5. Traverse each non-blank grid that is subdivided into a preset size. If the row and column boundary variable status in the current grid is (00000000) ₂ , mark the grid as an isolated pixel point; if the grid is For non-isolated pixels, go to step 6;

6. Break the grid of non-isolated pixels, mark the blank state of the rows and columns in the grid, connect the straight-line connected areas, and then determine the connected state of the grid based on the connection state of other boundaries and the center; if it is connected grid, go to step 9, if the connection state of the grid is uncertain, go to step 7;

7. Determine whether the boundary gaps are connected to each other by using the connected state of rows and columns, and determine the connected state of the grid; if it is a connected grid, perform step 9, and if the connected state of the grid is uncertain, perform step 8;

8. Use the method based on the region growing method to scan the state of 3 to 8 neighborhoods around the pixel point to obtain a connected grid;

9. Merge the obtained interconnected grids into the same set using tree-based set representation, count the set of grids, and the number of sets obtained is the connectivity of the image.

When processing an image, there are two types of points that affect image discrimination: one is the isolated point in the image, and the other is the missing pixel on the image. This type of point has an impact on the grid mark. For this type of point, In the technical solution of the present invention, filtering is performed by scanning pixels within a certain distance around (the particle size of the noise point is relatively small), since the possibility of isolated blank spots in the image is much smaller than the blank spots caused by noise, therefore, through this A method of filtering isolated blank points can effectively reduce the interference of noise on the calculation results. On the other hand, pixel areas are always larger than their boundaries, so in most cases, the number of single-pixel grids is much larger than that of non-single-pixel grids. Therefore, the calculation speed can be greatly improved by dividing the grid.

Due to the application of the above-mentioned technology, the present invention has the following advantages compared with the prior art:

1. The present invention can achieve a certain anti-noise effect by appropriately increasing the granularity in the identification process of identifying connectivity, and at the same time, the identification method can achieve an effect closer to human eye identification by appropriately enlarging the width of the connecting path.

2. Since the image is gridded step by step during the search, the method provided by the present invention has higher efficiency in searching connected regions than the unoptimized algorithm.

3. Since the present invention uses grid mark-based connectivity calculation to identify image features, it is highly sensitive to changes in image texture.

4. The present invention can realize non-marked search, so it is more likely to retain the original image for subsequent processing, which can effectively save the time and storage space requirements of the entire recognition process.

Description of drawings

Fig. 1 is the main flow of the identification application process provided by the embodiment of the present invention;

Fig. 2 is a comparison diagram of two different types of image effects in the recognition process using connectivity in the embodiment of the present invention;

Fig. 3 is a flow chart of a grid mark-based image recognition method provided by an embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with accompanying drawing and embodiment:

Example 1:

In this embodiment, the application of melon seeds screening is taken as an example, and multiple methods are used to jointly identify the images of melon seeds with mottled spots, so as to realize the intelligent screening of melon seeds.

Referring to accompanying drawing 1, it is the flow chart of the melon seed intelligent screening application process that the present embodiment provides. The dynamic capture image adopts Samsung SCC-C4203P low-illuminance color/black and white conversion industrial camera to obtain the melon seed array image on the assembly line and transmits it to the computer, and uses Microsoft Visual C++6.0 to preprocess the image transmitted to the computer. The eigenvalue extraction algorithm obtains the core eigenvalues, uses the judgment standard to screen out and mark the qualified melon seeds, and then sends a control signal to the connected mechanical equipment to screen out the qualified melon seeds.

When judging whether it is qualified, first use the long and short axis feature method, the pixel specific gravity method, and the histogram method to filter, and obtain the qualified melon seeds shown in the silhouette renderings of a1 to a4 in the accompanying drawing 2, but after filtering, there are still some seeds as shown in the accompanying drawing 2. The unqualified melon seeds shown in the b1~b4 silhouette renderings, that is, variegated melon seeds, can be seen from Figure 2. The size, size, black area coverage and histogram of the variegated melon seeds are similar to those of super-grade melon seeds, which cannot be obtained by existing methods. Sorting, but Huaban melon seeds also have their own unique characteristics - there are multiple circular patterns of different sizes on the surface. Therefore, the present invention uses the multi-connectivity of the calculation image to distinguish. Based on silhouette processing, a method of solving connectivity using set merging. In this embodiment, the criterion for judging is: the connectivity of premium melon seeds <= 3, and those greater than 3 are variegated melon seeds. In this way, petal melon seeds can be selected from standard melon seeds. Statistics show that the blank areas of the melon seed images obtained in the application are relatively small. Small, so the initial grid value set in this example can be small.

The melon seeds are binarized first, and then the shadow image of the melon seeds is obtained. After the two images are silhouetted, the multi-connectivity is calculated for the obtained silhouette image.

Referring to accompanying drawing 3, it is the flow chart of the image recognition method based on the grid label that the present embodiment provides, and its steps are:

Step 1: Forcibly set the boundary pixels of the image with a width of n as boundary points, and perform initialization operations such as allocating grid memory, constructing a linked list, and resetting the counter to zero.

Step 2: Scan each grid point in the image, and identify the pixel distribution parameters in each grid, including: ① byte-type row and column boundary variable status; ② index-type connected state variable mark; if the current grid row i is valid (not out of bounds), initialize the grid. If (1,1) is a boundary point, set its corresponding row and column positions to 0. First judge (1,3), (3,1), Whether (3,3) is a boundary point, if so, set the status to (00000000) ₂ and go to step 5; otherwise, continue to scan other pixels and go to step 3.

Step 3: If the current grid column j is valid (not out of bounds), go to step 5, otherwise move the grid down by one, and then go to step 2.

Step 4: Judge the status of the current grid.

Case1: If the status is (00000000) ₂ , set the mark of grid(i,j) to -1, and go to step 3.

Case2: If the status is (111111111) ₂ , first judge grid(i,j-1), whether the status of grid(i-1,j) is (111111111) ₂ . If so, compare the sets where these grids are located. If they are not in the same set, merge the sets, and then judge grid(i,j), grid(i,j-1), grid(i-1,j), grid( I-1, j-1) whether there are other n consecutive 1s in the area, if so, merge the grid it represents, and go to step 3; if not, then judge grid(i, j -1), grid(i-1,j), grid(i-1,j-1) whether there is a cell whose status is (00000000) ₂ , if so, go to step 3; if not, judge grid(i,j ), grid(i,j-1), grid(i-1,j), grid(i-1,j-1) whether the horizontal or vertical direction of the area A can be defined with grid(i,j ) intersecting straight line segments, if possible, mark all the corresponding status bits with continuous corresponding marks. If not, and there are consecutive n bits of 1 in the corresponding bit on the upper boundary or left boundary of A, merge the grids it represents and go to step 3.

Case3: When the status is in other cases, judge whether grid(i,j-1), grid(i-1,j), grid(i-1,j-1) has a status: (00000000) ₂ , If yes, go to step 3; if not, judge the area level composed of grid(i,j), grid(i,j-1), grid(i-1,j), grid(i-1,j-1) or Whether the vertical direction can pass through the defined line segment that intersects with grid(i, j), if so, mark all the corresponding status bits that have continuous corresponding 0; otherwise scan the pixel of this area to determine its Connectivity and merge the set of grids identified by regions adjacent to grid(i,j), grid(i,j-1), go to step 3.

Step 5: Count through the number of sets of linked lists, and the number of sets is the multi-connectivity. According to the judgment requirement to connectivity degree, set up figure, in the present embodiment is to determine the property of melon seeds.

The image recognition method provided by the invention improves the recognition efficiency on the premise of ensuring the recognition accuracy, is especially suitable for sorting and inspection of objects with different internal patterns recognized by a computer in large batches, and has a good application effect.

Claims (1)

1. A method for image recognition based on grid marks, characterized in that the image is set with the degree of connectivity of the image, comprising the steps: (1) Divide the image to be identified into several equal grids according to the granularity requirements, and pre-set the size of the grid. The minimum initial value of the grid size is 3×3 pixels; (2) Scan each grid point in the image, and identify the pixel distribution parameters in each grid, including: ① byte-type row and column boundary variable status; ② index-type connected state variable mark; (3) Merge the blank areas in the grid and record their connectivity status; (4) Divide the non-blank grid into smaller grids again, and return to step (2) if the grid size is still greater than the minimum value; (5) Traverse each non-blank grid that is subdivided into a preset size. If the row and column boundary variable status in the current grid is (00000000) ₂ , mark the grid as an isolated pixel point; if the grid For non-isolated pixels, perform step (6); (6) Break up the grid of non-isolated pixels, mark the blank state of the rows and columns in the grid, connect the straight-line connected areas, and then determine the connected state of the grid based on the connection state of other boundaries and the center; if it is connected For the grid, perform step (9), if the connectivity state of the grid is uncertain, perform step (7); (7) Determine whether the boundary gaps are connected to each other by using the connected state of rows and columns, and determine the connected state of the grid; if it is a connected grid, perform step (9); if the connected state of the grid is uncertain, perform step (8); (8) Use the method based on the region growing method to scan the state of the neighborhood around the pixel point to obtain a connected grid; (9) Merge the obtained interconnected grids into the same set using tree-based set representation, count the set of grids, and the number of sets obtained is the connectivity of the image.

Images